Agent Data Tools: AI MCP server
io.github.Nutshell2000/agent-data-tools-aiNutshell2000streamable-httpv1.331.0Scanned
33 tools: LLM chat, summaries, translation, embeddings, speech, images. From $0.002.
33 tools: 5 do something that can't be undone, 25 change things and 3 only read.
- Reads private data · 1
- Reads text others write · 33
- Can send data out · 11
- 6 findings
Tools listed without sign-in. Remote posture
With Anlyon
5 tools here do something that can't be undone. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the Agent Data Tools: AI MCP server can do
The Agent Data Tools: AI MCP server is published by the GitHub account Nutshell2000. It has 33 tools. It can do things that cannot be undone, such as sending, paying or publishing (ai_chat, ai_chat_large, ai_translate, dub and 1 more) and create or change data (25 tools: ai_embed, transcribe, ai_speech, image and 21 more), but cannot delete data. 3 tools only read.
1 tool reads private data (transcribe_summary); 33 tools bring in text other people can write (ai_chat, ai_chat_large, ai_embed and 30 more); 11 tools can send data out of your system (ai_chat, ai_chat_large, transcribe and 8 more). On its own it has all three parts of a prompt-injection exfiltration path: private data, text others write, and a way out.
The fixed checks matched 6 findings (annotation-conflict). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed 10 times since Anlyon Radar first saw them.
Last scanned Nov 21, 2026, version 1.331.0.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
33 tools
Can't be undone 5Claude Code: ask
ai_chatSend a prompt, get an answer from Meta's Llama 4 Scout (17B active, 16 experts), with an optional system prompt, up to 1,024 output tokens and adjustable temperature.Reads text others writeCan send data out1 finding
Why it is in this group
- The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe description's first verb is "send": it sends or publishes something to other people.
Findings
- Warn
The server marks "ai_chat" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
Description, as the server wrote it Send a prompt, get an answer from Meta's Llama 4 Scout (17B active, 16 experts), with an optional system prompt, up to 1,024 output tokens and adjustable temperature. Fast and cheap. Use for drafting, reasoning steps, Q&A and any quick LLM call without an AI provider account; other open models by name at /v1/chat/completions, harder reasoning at ai_chat_large. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.002/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Chat completions (LLM): Llama 4 Scout
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_chat_largeSend a prompt, get an answer from a larger open model: Llama 3.3 70B Instruct (default) or OpenAI gpt-oss-120b, with an optional system prompt, up to 1,024 output tokens and adjust…Reads text others writeCan send data out1 finding
Why it is in this group
- The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe description's first verb is "send": it sends or publishes something to other people.
Findings
- Warn
The server marks "ai_chat_large" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
Description, as the server wrote it Send a prompt, get an answer from a larger open model: Llama 3.3 70B Instruct (default) or OpenAI gpt-oss-120b, with an optional system prompt, up to 1,024 output tokens and adjustable temperature. Same call as /ai/chat with a stronger model. Use when Llama 4 Scout is not enough: reasoning, drafting and Q&A. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: LLM inference, large model: Llama 3.3 70B or gpt-oss-120b
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_translateTranslate text into another language with OpenAI's open-weight gpt-oss-20b, detecting the source language.Reads text others writeCan send data out1 finding
Why it is in this group
- The input addresses people (input.to): what it sends cannot be taken back.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a destination (input.to): what the tool writes goes there.
Findings
- Warn
The server marks "ai_translate" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
Description, as the server wrote it Translate text into another language with OpenAI's open-weight gpt-oss-20b, detecting the source language. Long text is split at sentence ends and checked, so a cut-off translation is an uncharged 502, never a success. Give the target as an ISO code such as en, sv, de, fr, es, zh or ja; less common languages use Meta's M2M100. Use to localise messages and documents. Example: text=Payment received, thank you!, to=sv. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Translate text, translation API: translate text into another language, about 100 languages
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
dubDub a clip: speech in any of 90+ languages is transcribed, translated and spoken again, with translated SRT on the original timings.Reads text others writeCan send data out1 finding
Why it is in this group
- The input addresses people (input.to): what it sends cannot be taken back.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a destination (input.to): what the tool writes goes there.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Findings
- Warn
The server marks "dub" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
Description, as the server wrote it Dub a clip: speech in any of 90+ languages is transcribed, translated and spoken again, with translated SRT on the original timings. New voice in English or Spanish (Aura-2 MP3, pick a voice, up to 2,500 characters); fr, zh, ja and ko are retired with MeloTTS (Workers AI, 20 Nov 2026). MP3/WAV up to 10 minutes or other formats up to 2 MB, by URL or upload. Keeps the order, not exact timing. Use for course clips and product videos. Example: url=https://upload.wikimedia.org/wikipedia/commons/7/77/Tofino_pronunciation.mp3, to=es. Price $0.10/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Dubbing, dub audio: translated voice track in 6 languages, plus subtitles
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
subtitles_translateTranslate an SRT or WebVTT subtitle file into another language cue by cue, keeping every timing and the file format, so the result drops straight into a video player or editor.Reads text others writeCan send data out1 finding
Why it is in this group
- The input addresses people (input.to): what it sends cannot be taken back.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a destination (input.to): what the tool writes goes there.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Findings
- Warn
The server marks "subtitles_translate" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
Description, as the server wrote it Translate an SRT or WebVTT subtitle file into another language cue by cue, keeping every timing and the file format, so the result drops straight into a video player or editor. Up to 400 cues (40,000 characters) as text or by URL; the source language is detected unless given. Meta M2M100 translation, 100 languages. Use to subtitle videos, courses and podcasts in more languages; pair with transcribe_long, which writes the first SRT. Price $0.02/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Translate subtitles: SRT or VTT into another language, timings kept
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
Changes things 25Claude Code: ask
ai_answerAnswer a question from the live web in one call: a Brave web search, the top result pages read, and an answer written by Llama 4 Scout from those sources only, with inline [n] cita…Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Answer a question from the live web in one call: a Brave web search, the top result pages read, and an answer written by Llama 4 Scout from those sources only, with inline [n] citations and the source list (title, URL). Says so when the sources do not answer it. Results are for the caller's immediate use only: no storing, resale or model training (Brave API terms). Use when an agent needs a current, sourced answer rather than raw search results. Example: q=What is the x402 payment protocol?. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Web answer, AI with citations: answer a question from live search results
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_classifyZero-shot classification: give text and 2-20 candidate labels, get the best label, a short reason and a ranking.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Zero-shot classification: give text and 2-20 candidate labels, get the best label, a short reason and a ranking. Use for routing tickets, tagging content, intent detection and moderation queues. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Text classification into labels: sort text into one of your labels (zero-shot), for routing and tagging
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_embedA 1,024-dimension embedding vector for up to 4,000 characters of text in 100+ languages, from BAAI's BGE-M3.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it A 1,024-dimension embedding vector for up to 4,000 characters of text in 100+ languages, from BAAI's BGE-M3. This tool answers in its own shape (dimensions, embedding, model); the OpenAI embeddings shape is at /v1/embeddings. Use for semantic search, RAG retrieval, clustering and deduplication. Example: text=pay-per-call APIs for AI agents. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: OpenAI compatible embeddings: BGE-M3, 1,024 dimensions
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_extract_jsonPull named fields out of unstructured text (emails, invoices, listings, bios) and get them back as a JSON object, with null for anything not present.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Pull named fields out of unstructured text (emails, invoices, listings, bios) and get them back as a JSON object, with null for anything not present. Use to turn messy text into structured data for an agent. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Extract JSON fields from text: turn unstructured text into a JSON object with the fields you name
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_fact_checkCheck a claim against the live web in one call: a Brave search, the top pages read, and a verdict from Llama 4 Scout using those sources only (supported, contradicted, mixed or unv…Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Check a claim against the live web in one call: a Brave search, the top pages read, and a verdict from Llama 4 Scout using those sources only (supported, contradicted, mixed or unverified) with a short explanation, [n] citations and the sources. A model's reading of a few search results, not an editorial fact-check. Results are for immediate use only (Brave API terms). Use to screen a statement before an agent repeats or acts on it. Example: claim=The x402 protocol uses the HTTP 402 Payment Required status code.. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Fact check, verify a claim: against live web sources
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_keywordsThe most important keywords and key phrases of a text, most relevant first.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it The most important keywords and key phrases of a text, most relevant first. Use for tagging, SEO, search indexing and topic detection. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Keyword extraction: pull the main keywords and key phrases out of a text
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_model_gpt_oss_120bChat with OpenAI's open-weight gpt-oss-120b reasoning model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output…Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Chat with OpenAI's open-weight gpt-oss-120b reasoning model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens. Also reachable as model "gpt-oss-120b" on the OpenAI-compatible /v1/chat/completions. No account or provider key. Use when an agent wants this model by name. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: GPT-OSS 120B chat API: pay per call, no API key
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_model_gpt_oss_20bChat with OpenAI's open-weight gpt-oss-20b reasoning model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output …Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Chat with OpenAI's open-weight gpt-oss-20b reasoning model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens. Also reachable as model "gpt-oss-20b" on the OpenAI-compatible /v1/chat/completions. No account or provider key. Use when an agent wants this model by name. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.002/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: GPT-OSS 20B chat API: pay per call, no API key
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_model_mistral_smallChat with Mistral AI's Mistral Small 3.1 24B model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Chat with Mistral AI's Mistral Small 3.1 24B model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens. Also reachable as model "mistral-small" on the OpenAI-compatible /v1/chat/completions. No account or provider key. Use when an agent wants this model by name. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.005/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Mistral Small 3.1 chat API: pay per call, no API key
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_model_qwen_coderChat with Alibaba's Qwen2.5 Coder 32B code model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Chat with Alibaba's Qwen2.5 Coder 32B code model (Apache 2.0) through one paid HTTP call: send a prompt and an optional system prompt, get the answer, up to 1,024 output tokens. Also reachable as model "qwen-coder" on the OpenAI-compatible /v1/chat/completions. No account or provider key. Use when an agent wants this model by name. Example: prompt=Explain HTTP status code 402 in one sentence.. Price $0.005/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Qwen2.5 Coder 32B, code LLM API: pay per call, no API key
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_moderationClassify a text as safe or unsafe with Meta's Llama Guard 3 and name the hazard categories it falls into (violent and non-violent crimes, sexual content, child exploitation, hate, …Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Classify a text as safe or unsafe with Meta's Llama Guard 3 and name the hazard categories it falls into (violent and non-violent crimes, sexual content, child exploitation, hate, self-harm, weapons, privacy, defamation, elections and others; 14 in all). Up to 8,000 characters. Also available in the OpenAI moderation shape at /v1/moderations. Use to screen user messages, model outputs or content before an agent posts or acts on it. Example: text=How do I bake sourdough bread at home?. Price $0.002/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Content moderation, safety check: Llama Guard 3
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_outlineA structured Markdown outline (headings and bullet points) of a text you send, or a suggested outline for a topic.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it A structured Markdown outline (headings and bullet points) of a text you send, or a suggested outline for a topic. Use to plan articles, reports, talks and documentation. Example: text=How small businesses can accept stablecoin payments. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Outline a text or topic: make a structured Markdown outline of a text or topic
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_rerankOrder up to 50 documents by how well each answers a query, with a relevance score from 0 to 1 (BAAI bge-reranker-base, a cross-encoder).Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Order up to 50 documents by how well each answers a query, with a relevance score from 0 to 1 (BAAI bge-reranker-base, a cross-encoder). Send documents as a JSON array or one per line; each is read up to 2,000 characters, 40,000 in all. Returns the original index, score and text, best first; top_n keeps only the best. Also available in the Cohere-style shape at /v1/rerank. Use to pick what a model reads after search or vector retrieval. Price $0.002/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Rerank documents by relevance: reranker for RAG
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_rewriteRewrite text to be more formal, casual, simpler, shorter, friendlier, more professional or more persuasive, keeping the meaning and language.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Rewrite text to be more formal, casual, simpler, shorter, friendlier, more professional or more persuasive, keeping the meaning and language. Use to polish emails, replies, product copy and messages. Example: text=hey, cant make it tmrw, can we move to thu?. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Rewrite text in a style: rewrite text to be more formal, casual, simpler, shorter or more persuasive
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_sentimentPositive or negative sentiment of English text with a confidence score, from a DistilBERT model fine-tuned on SST-2.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Positive or negative sentiment of English text with a confidence score, from a DistilBERT model fine-tuned on SST-2. Use to triage reviews, support messages, social posts and replies. Example: text=The checkout was fast and support answered within minutes.. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Sentiment analysis of text: tell whether a text is positive or negative, with a confidence score
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
ai_speechTurn up to 300 characters of text into natural, human-sounding speech: MP3 in English (40 voices) or Spanish (10 voices) with Deepgram Aura-2.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Turn up to 300 characters of text into natural, human-sounding speech: MP3 in English (40 voices) or Spanish (10 voices) with Deepgram Aura-2. fr, zh, ja and ko are retired with MeloTTS (Workers AI, 20 Nov 2026) and are refused uncharged; use en or es. Pick a voice by name or leave the default. Use for voice replies, audio notes, accessibility and phone or chat bots; for up to 2,000 characters use text_to_speech. Example: text=Your payment was received. Thank you!. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Text to speech (TTS): natural voices, audio file
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
dialogueTurn a script into a multi-voice conversation: write one line per turn as "Name: text", each speaker gets their own natural voice (Deepgram Aura-2, English or Spanish, or pick voic…Reads text others write1 finding
Why it is in this group
- The input takes a statement or command to run (input.script), which can change anything it can reach.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Warn
The server marks "dialogue" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input takes a statement or command to run (input.script), which can change anything it can reach.
Description, as the server wrote it Turn a script into a multi-voice conversation: write one line per turn as "Name: text", each speaker gets their own natural voice (Deepgram Aura-2, English or Spanish, or pick voices), and the whole dialogue comes back as one MP3 with SRT captions naming the speaker, timed from the real audio. Up to 6 speakers and 2,500 characters. Use for podcast intros, language-lesson dialogues, explainer videos, ads and game or story scenes. Example: script=Anna: Welcome back to the show! Ben: Thanks, it is great to be here.. Price $0.10/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Dialogue to audio: a script with several speakers as one MP3, plus captions
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
imageText-to-image generation with FLUX.1 schnell.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Text-to-image generation with FLUX.1 schnell. Send a prompt, get a 1024x1024 JPEG back as base64 in a few seconds. Use for illustrations, thumbnails, mockups, avatars and social media images. Optional steps parameter (1-8, default 4) trades speed for detail. Example: prompt=a lighthouse on a cliff at sunset, watercolor. Price $0.03/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Image generation, generate image from a text prompt: create a picture from a description with an AI image model
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
text_to_speechText to speech for a paragraph or a page: up to 2,000 characters as one MP3 in English (40 voices) or Spanish (10 voices) with Deepgram Aura-2; fr, zh, ja and ko are retired with M…Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Text to speech for a paragraph or a page: up to 2,000 characters as one MP3 in English (40 voices) or Spanish (10 voices) with Deepgram Aura-2; fr, zh, ja and ko are retired with MeloTTS (Workers AI, 20 Nov 2026). Pick a voice by name or leave the default. Use for narration, article and summary read-outs, voice replies, IVR prompts and accessibility; for a sentence of up to 300 characters ai_speech costs less. Price $0.08/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Text to speech: long text, up to 2,000 characters as one MP3, natural voices
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
text_to_speech_captionsText to speech that comes with subtitles: up to 2,000 characters spoken as one MP3 in English or Spanish (Deepgram Aura-2, pick a voice), plus SRT or VTT captions whose timings com…Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Text to speech that comes with subtitles: up to 2,000 characters spoken as one MP3 in English or Spanish (Deepgram Aura-2, pick a voice), plus SRT or VTT captions whose timings come from the generated audio itself, sentence by sentence. Ready for a video editor, no transcription step. Use for narrated videos, social clips, e-learning and accessible audio with captions. Price $0.08/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Text to speech with subtitles: one MP3 plus SRT or VTT timed to the voice
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
text_to_speech_liteLow-cost text to speech for longer English text: up to 2,000 characters as one MP3 (base64) with Deepgram Aura-1, 12 voices.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Low-cost text to speech for longer English text: up to 2,000 characters as one MP3 (base64) with Deepgram Aura-1, 12 voices. Half the price per character of text_to_speech, with a slightly less natural voice. Pick a voice by name or leave the default. Use for narration drafts, read-aloud of articles and summaries, notifications and accessibility at volume; for the most natural voices or Spanish use text_to_speech. Example: text=Your order has shipped and should reach you within three working days.. Price $0.04/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Cheap text to speech: up to 2,000 characters in English as one MP3
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
transcribeTranscribe speech from an audio file at a public URL, or uploaded in a POST body, using Whisper large-v3-turbo.Reads text others writeCan send data outNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Transcribe speech from an audio file at a public URL, or uploaded in a POST body, using Whisper large-v3-turbo. Returns the transcript text, detected language, duration and word count; subtitles=srt or vtt adds subtitle text and timestamps=true adds timed segments and words. Accepts common formats such as MP3, WAV, M4A and OGG up to 5 MB, in 90+ languages. Use for voice notes, podcast clips, meeting snippets, subtitles and voice messages. Example: url=https://commons.wikimedia.org/wiki/Special:FilePath/En-us-hello.ogg. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Speech to text, transcribe audio: Whisper transcription of an audio file
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
transcribe_longTranscribe a podcast, meeting, interview or lecture: up to 60 minutes of MP3 or WAV (25 MB) by URL or uploaded in a POST body, 90+ languages, Whisper large-v3-turbo.Reads text others writeCan send data outNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Transcribe a podcast, meeting, interview or lecture: up to 60 minutes of MP3 or WAV (25 MB) by URL or uploaded in a POST body, 90+ languages, Whisper large-v3-turbo. Parts are joined, so subtitles=srt or vtt and timestamps=true cover the whole recording; longer audio is refused before anything runs. Use for meeting notes, podcast show notes, video subtitles and interviews; for a short clip in any format, transcribe costs less. Example: url=https://upload.wikimedia.org/wikipedia/commons/7/77/Tofino_pronunciation.mp3. Price $0.05/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Whisper audio transcription: up to 60 minutes of MP3 or WAV, with subtitles
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
transcribe_speakersTranscribe a meeting, interview, call or podcast with speaker labels: each turn says which speaker talked, with start and end times (Deepgram Nova-3).Reads text others writeCan send data outNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it Transcribe a meeting, interview, call or podcast with speaker labels: each turn says which speaker talked, with start and end times (Deepgram Nova-3). MP3/WAV up to 15 minutes, other formats up to 2 MB, by URL or upload. Optional redaction of personal data (pii), card numbers (pci) or numbers, and subtitles with the speaker in each line. Use for meeting minutes, interview transcripts, call reviews and podcast show notes. Example: url=https://upload.wikimedia.org/wikipedia/commons/7/77/Tofino_pronunciation.mp3. Price $0.10/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Speaker diarization and labels: transcribe audio with who said what
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
voice_agentOne turn of a voice assistant: send the user's voice note (or text), your instructions and the earlier turns; get back what was heard, a short reply, the reply spoken in English or…Reads text others writeCan send data outNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Description, as the server wrote it One turn of a voice assistant: send the user's voice note (or text), your instructions and the earlier turns; get back what was heard, a short reply, the reply spoken in English or Spanish and the updated history to send next time. Speech up to 2 minutes (MP3/WAV, or 1 MB of other formats); replies up to 600 characters. Use for voice bots in Telegram or WhatsApp, phone menus, kiosks and hands-free helpers. Example: text=What time do you open on Saturday?. Price $0.05/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Voice agent: spoken question in, spoken answer out, with memory
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
Only reads 3Show 3 read-only tools
ai_summarize_textSummarize up to 12,000 characters of text you send into a few sentences plus bullet key points.Reads text others writeNo findings
Why it is in this group
- The name "ai_summarize_text" says "summarize": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Summarize up to 12,000 characters of text you send into a few sentences plus bullet key points. Use for long emails, documents, transcripts and tool outputs (for a web page by URL, use /summarize). Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Summarize text: shorten long text into a few sentences and key points
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
describeVision model for an image at a public URL: get a detailed description including any visible text, or pass a question to ask something specific about the image.Reads text others writeCan send data outNo findings
Why it is in this group
- The name "describe" says "describe": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "url": it reads web pages, which anyone can write.
- untrusted-inputThe input takes a URL (input.url), and the tool reads it: it reads web pages, which anyone can write.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Claude Code: ask, because it is tagged outbound.
Description, as the server wrote it Vision model for an image at a public URL: get a detailed description including any visible text, or pass a question to ask something specific about the image. Accepts PNG, JPEG, WebP and GIF up to 3 MB; a web page URL is read through its preview image (og:image). Use for alt text, OCR-style reading of screenshots, chart reading and content checks. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Describe image, ask about it: ask what is in an image or photo at a URL and get a text description or answer
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
transcribe_summarySummarize a voice note, meeting, call or podcast in one call: transcribes up to 60 minutes (MP3/WAV up to 25 MB; M4A, OGG or WebM up to 5 MB) by URL or upload, then returns a title…Reads private dataReads text others writeCan send data outNo findings
Why it is in this group
- The description's first verb is "summarize": a word for reading.
What it can reach
- private-dataThe description's first sentence says "summarize" and "note": it reads notes.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "summarize" and "web": it reads web pages, which anyone can write.
- untrusted-inputThe input takes a URL (input.url), and the tool reads it: it reads web pages, which anyone can write.
- outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
Claude Code: ask, because it is tagged outbound.
Description, as the server wrote it Summarize a voice note, meeting, call or podcast in one call: transcribes up to 60 minutes (MP3/WAV up to 25 MB; M4A, OGG or WebM up to 5 MB) by URL or upload, then returns a title, summary, key points, action items with owner and due date, open questions and the transcript, in the recording's language. Use for meeting notes, voicemail to CRM and call notes. Example: url=https://upload.wikimedia.org/wikipedia/commons/7/77/Tofino_pronunciation.mp3. Price $0.05/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.
Annotations the server declares
- title: Meeting transcript and summary: voice note or meeting recording to summary and action items
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
363 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 1 private-data tool breaks every one.
- Reads private data1
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write33
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out11
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
{
"permissions": {
"deny": [
"mcp__agent-data-tools-ai__transcribe_summary"
]
}
}Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:59 UTC, unchanged since Nov 20, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 30
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- https://agent-data-tools.revmesh2074.workers.dev/.well-known/oauth-protected-resource
- Authorization servers
- https://agent-data-tools.revmesh2074.workers.dev
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- yes
- Client ID metadata documents
- no
- HighThe whole server
The endpoint served tools/list without credentials, and the list has 30 tools that change things (5 irreversible, 25 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://agent-data-tools.revmesh2074.workers.dev/mcp/ai: tools/list returned 33 tools to an anonymous client, including ai_chat (irreversible), ai_chat_large (irreversible), ai_translate (irreversible), dub (irreversible), subtitles_translate (irreversible) and 25 more
Posture changes
- Nov 20, 2026
- Tools that change things, listed without credentials: 31 to 30
- Nov 13, 2026
- Tools that change things, listed without credentials: 23 to 31
- Oct 29, 2026
- Tools that change things, listed without credentials: 13 to 23
Findings
Matches of fixed checks, with the text they matched.
- Warnai_chat
The server marks "ai_chat" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
- Warnai_chat_large
The server marks "ai_chat_large" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
- Warnai_translate
The server marks "ai_translate" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
- Warndub
The server marks "dub" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
The server marks "subtitles_translate" read-only (readOnlyHint: true), and its own name, description or input give the class "irreversible".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input addresses people (input.to): what it sends cannot be taken back.
- Warndialogue
The server marks "dialogue" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The input takes a statement or command to run (input.script), which can change anything it can reach.
History
Changes to its tool definitions and to its remote endpoint, newest first.
Tools12 changed.~12 changedView diff
- Changed:ai_chatCan't be undonedescription
- Changed:ai_classifyChanges thingsdescription
- Changed:ai_embedChanges thingsdescription
- Changed:ai_extract_jsonChanges thingsdescription
- Changed:ai_keywordsChanges thingsdescription
- Changed:ai_moderationChanges thingsdescription
- Changed:ai_outlineChanges thingsdescription
- Changed:ai_rerankChanges thingsdescription
- Changed:ai_rewriteChanges thingsdescription
- Changed:ai_sentimentChanges thingsdescription
- Changed:ai_summarize_textOnly readsdescription
- Changed:ai_translateCan't be undonedescription
Tools2 changed.~2 changedView diff
- Changed:ai_speechChanges thingsdescription
- Changed:text_to_speechChanges thingsdescription
PostureRemote posture changed. Tools that change things, listed without credentials: 31 -> 30.View diff
- Tools that change things, listed without credentials: 31 to 30
Tools2 removed, 1 changed.−2 removed~1 changedView diff
- Removed:ai_model_deepseek_r1
- Removed:ai_model_qwen3_30b
- Changed:ai_speechChanges thingsdescription
Tools6 changed.~6 changedView diff
- Changed:ai_model_gpt_oss_20bChanges thingsdescription
- Changed:ai_model_mistral_smallChanges thingsdescription
- Changed:ai_model_qwen3_30bChanges thingsdescription
- Changed:ai_model_qwen_coderChanges thingsdescription
- Changed:ai_moderationChanges thingsdescription
- Changed:ai_rerankChanges thingsdescription
Tools4 changed.~4 changedView diff
- Changed:ai_speechChanges thingsinputSchema
- Changed:dubCan't be undonedescriptioninputSchema
- Changed:text_to_speechChanges thingsdescriptioninputSchema
- Changed:voice_agentChanges thingsdescriptioninputSchema
Tools2 changed.~2 changedView diff
- Changed:ai_chatCan't be undoneinputSchema
- Changed:transcribeChanges thingsinputSchema
Tools2 changed.~2 changedView diff
- Changed:ai_speechChanges thingsinputSchema
- Changed:text_to_speechChanges thingsinputSchema
Tools9 removed, 12 changed.−9 removed~12 changedView diff
- Removed:ai_embeddings
- Removed:ai_sentiment_analysis
- Removed:ai_summarize
- Removed:ai_translate_text
- Removed:ai_tts
- Removed:audio_transcribe
- Removed:image_generate
- Removed:llm_chat
- Removed:llm_inference
- Changed:ai_chatCan't be undoneinputSchema
- Changed:ai_chat_largeCan't be undoneinputSchema
- Changed:ai_keywordsChanges thingsinputSchema
- Changed:ai_model_deepseek_r1Changes thingsinputSchema
- Changed:ai_model_gpt_oss_120bChanges thingsinputSchema
- Changed:ai_model_gpt_oss_20bChanges thingsinputSchema
- Changed:ai_model_mistral_smallChanges thingsinputSchema
- Changed:ai_model_qwen3_30bChanges thingsinputSchema
- Changed:ai_model_qwen_coderChanges thingsinputSchema
- Changed:ai_outlineChanges thingsinputSchema
- Changed:ai_summarize_textOnly readsinputSchema
- Changed:imageChanges thingsinputSchema
Tools4 changed.~4 changedView diff
- Changed:dubCan't be undonetitleannotations
- Changed:transcribe_longChanges thingstitleannotations
- Changed:transcribe_speakersChanges thingstitleannotations
- Changed:transcribe_summaryOnly readstitleannotations
Install
Pinned to version 1.331.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http agent-data-tools-ai https://agent-data-tools.revmesh2074.workers.dev/mcp/aiclaude mcp add --transport http agent-data-tools-ai https://agent-data-tools.revmesh2074.workers.dev/mcp/aiRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__agent-data-tools-ai__ai_summarize_text"
],
"ask": [
"mcp__agent-data-tools-ai__ai_chat",
"mcp__agent-data-tools-ai__ai_chat_large",
"mcp__agent-data-tools-ai__ai_embed",
"mcp__agent-data-tools-ai__transcribe",
"mcp__agent-data-tools-ai__ai_speech",
"mcp__agent-data-tools-ai__image",
"mcp__agent-data-tools-ai__describe",
"mcp__agent-data-tools-ai__ai_translate",
"mcp__agent-data-tools-ai__ai_extract_json",
"mcp__agent-data-tools-ai__ai_answer",
"mcp__agent-data-tools-ai__ai_fact_check",
"mcp__agent-data-tools-ai__text_to_speech",
"mcp__agent-data-tools-ai__ai_rerank",
"mcp__agent-data-tools-ai__ai_moderation",
"mcp__agent-data-tools-ai__ai_model_gpt_oss_20b",
"mcp__agent-data-tools-ai__ai_model_gpt_oss_120b",
"mcp__agent-data-tools-ai__ai_model_mistral_small",
"mcp__agent-data-tools-ai__ai_model_qwen_coder",
"mcp__agent-data-tools-ai__ai_sentiment",
"mcp__agent-data-tools-ai__ai_classify",
"mcp__agent-data-tools-ai__ai_keywords",
"mcp__agent-data-tools-ai__ai_rewrite",
"mcp__agent-data-tools-ai__ai_outline",
"mcp__agent-data-tools-ai__transcribe_long",
"mcp__agent-data-tools-ai__transcribe_summary",
"mcp__agent-data-tools-ai__text_to_speech_lite",
"mcp__agent-data-tools-ai__voice_agent",
"mcp__agent-data-tools-ai__dub",
"mcp__agent-data-tools-ai__transcribe_speakers",
"mcp__agent-data-tools-ai__subtitles_translate",
"mcp__agent-data-tools-ai__dialogue",
"mcp__agent-data-tools-ai__text_to_speech_captions"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as agent-data-tools-ai. Review it before you use it.
Capability tags behind 33 rules
- mcp__agent-data-tools-ai__ai_chat untrusted-input, outbound
- mcp__agent-data-tools-ai__ai_chat_large untrusted-input, outbound
- mcp__agent-data-tools-ai__ai_embed untrusted-input
- mcp__agent-data-tools-ai__transcribe untrusted-input, outbound
- mcp__agent-data-tools-ai__ai_speech untrusted-input
- mcp__agent-data-tools-ai__image untrusted-input
- mcp__agent-data-tools-ai__describe untrusted-input, outbound
- mcp__agent-data-tools-ai__ai_summarize_text untrusted-input
- mcp__agent-data-tools-ai__ai_translate untrusted-input, outbound
- mcp__agent-data-tools-ai__ai_extract_json untrusted-input
- mcp__agent-data-tools-ai__ai_answer untrusted-input
- mcp__agent-data-tools-ai__ai_fact_check untrusted-input
- mcp__agent-data-tools-ai__text_to_speech untrusted-input
- mcp__agent-data-tools-ai__ai_rerank untrusted-input
- mcp__agent-data-tools-ai__ai_moderation untrusted-input
- mcp__agent-data-tools-ai__ai_model_gpt_oss_20b untrusted-input
- mcp__agent-data-tools-ai__ai_model_gpt_oss_120b untrusted-input
- mcp__agent-data-tools-ai__ai_model_mistral_small untrusted-input
- mcp__agent-data-tools-ai__ai_model_qwen_coder untrusted-input
- mcp__agent-data-tools-ai__ai_sentiment untrusted-input
- mcp__agent-data-tools-ai__ai_classify untrusted-input
- mcp__agent-data-tools-ai__ai_keywords untrusted-input
- mcp__agent-data-tools-ai__ai_rewrite untrusted-input
- mcp__agent-data-tools-ai__ai_outline untrusted-input
- mcp__agent-data-tools-ai__transcribe_long untrusted-input, outbound
- mcp__agent-data-tools-ai__transcribe_summary private-data, untrusted-input, outbound
- mcp__agent-data-tools-ai__text_to_speech_lite untrusted-input
- mcp__agent-data-tools-ai__voice_agent untrusted-input, outbound
- mcp__agent-data-tools-ai__dub untrusted-input, outbound
- mcp__agent-data-tools-ai__transcribe_speakers untrusted-input, outbound
- mcp__agent-data-tools-ai__subtitles_translate untrusted-input, outbound
- mcp__agent-data-tools-ai__dialogue untrusted-input
- mcp__agent-data-tools-ai__text_to_speech_captions untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/io.github.Nutshell2000/agent-data-tools-ai/verify?hash=ca8696bd81b6cf542811e03f92132681081ae34a35fd8121cdc7a1f334159974" | jq -e '.changed == false'Pinned hash ca8696bd81b6, from Nov 21, 2026, 19:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.Nutshell2000/agent-data-tools-ai/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.Nutshell2000/agent-data-tools-ai)Source: MCP Registry · repository. Synced Nov 21, 2026, 19:17 UTC.
Agent Data Tools: AI MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Agent Data Tools: AI MCP server safe?
- No scan can promise that. What its definitions show: 33 tools, 30 of which can change, send or delete something, and 6 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the Agent Data Tools: AI MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the Agent Data Tools: AI MCP server do things that cannot be undone?
- Yes. 5 tools can send, pay, publish or merge: ai_chat, ai_chat_large, ai_translate, dub and subtitles_translate.
What tools does the Agent Data Tools: AI MCP server have?
- 33 tools: 5 irreversible, 25 write, 3 read. ai_chat, ai_chat_large, ai_translate, dub, subtitles_translate, ai_embed, transcribe, ai_speech and 25 more.
Does the Agent Data Tools: AI MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the Agent Data Tools: AI MCP server send data to outside services?
- 11 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): ai_chat, ai_chat_large, transcribe, describe, ai_translate, transcribe_long and 5 more.
Has the Agent Data Tools: AI MCP server changed its tools?
- Yes, 10 times since Anlyon Radar started tracking it, most recently on Nov 21, 2026. Every change is on this page and in its Atom feed.