Skip to content

Agent Toolbox MCP server

io.github.zaffy18/agent-toolboxzaffy18streamable-httpv1.0.0Scanned

Everyday tools for AI agents, paid per call with x402: images, markdown, speech-to-text, embeddings.

24 tools: 1 can delete or cancel things, 16 change things and 7 only read.

Tools: 1 delete, 0 irreversible, 16 write, 7 read.
  • Reads private data · 2
  • Reads text others write · 21
  • Can send data out · 7
  • 4 findings

Tools listed without sign-in. Remote posture

With Anlyon

1 tool here can delete or cancel things. Govern it.

  • Anlyon lists the tools and adopts none.
  • Each tool you adopt gets approvals and receipts.
  • Your agent never holds the key.
Govern this server →

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 Toolbox MCP server can do

The Agent Toolbox MCP server is published by the GitHub account zaffy18. It has 24 tools. It can delete or cancel things (1 tool: stop_watch) and create or change data (16 tools: generate_image, file_to_markdown, transcribe_audio, embed_text and 12 more). 7 tools only read.

2 tools read private data (url_to_markdown and watch_status); 21 tools bring in text other people can write (generate_image, url_to_markdown, file_to_markdown and 18 more); 7 tools can send data out of your system (url_to_markdown, summarize, extract_data and 4 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 4 findings (annotation-conflict). Each one comes with the text it matched.

Its remote endpoint lists its tools without sign-in, and its tools changed once since Anlyon Radar first saw them.

Last scanned Nov 9, 2026, version 1.0.0.

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.

24 tools

Deletes or cancels 1Claude Code: deny
  • stop_watchFree.No findings

    Why it is in this group

    • The name "stop_watch" says "stop": a word for a change.
    • The server marks it destructive (destructiveHint: true).
    Description, as the server wrote it
    Free. Stop a field watch, with the watch_id and secret that watch_field returned.

    Annotations the server declares

    • title: Stop Watch
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: false
Changes things 16Claude Code: ask
  • chat_gpt_oss_120bOpenAI-compatible chat completion with gpt-oss-120b (OpenAI's open-weight reasoning model): set the base URL to /v1/llm/gpt-oss-120b in any OpenAI SDK.Reads text others write1 finding

    Why it is in this group

    • The description's first verb is "open": a word for a change.
    • 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 "chat_gpt_oss_120b" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

      Matched text

      annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "open": a word for a change.

      What this check looks for
    Description, as the server wrote it
    OpenAI-compatible chat completion with gpt-oss-120b (OpenAI's open-weight reasoning model): set the base URL to /v1/llm/gpt-oss-120b in any OpenAI SDK. Up to 32,000 characters in and 2,048 tokens out per call, a fixed price per call, no streaming. Costs USD 0.005 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Chat Gpt Oss 120b
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • chat_qwen3_30bOpenAI-compatible chat completion with Qwen3 30B (fast mixture-of-experts model): set the base URL to /v1/llm/qwen3-30b in any OpenAI SDK.Reads text others write1 finding

    Why it is in this group

    • The description's first verb is "open": a word for a change.
    • 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 "chat_qwen3_30b" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

      Matched text

      annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "open": a word for a change.

      What this check looks for
    Description, as the server wrote it
    OpenAI-compatible chat completion with Qwen3 30B (fast mixture-of-experts model): set the base URL to /v1/llm/qwen3-30b in any OpenAI SDK. Up to 32,000 characters in and 2,048 tokens out per call, a fixed price per call, no streaming. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Chat Qwen3 30b
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • company_leiFull official record of a legal entity by its LEI (Legal Entity Identifier) from GLEIF: legal and other names, registered and head-office addresses, company-register number, legal …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
    Full official record of a legal entity by its LEI (Legal Entity Identifier) from GLEIF: legal and other names, registered and head-office addresses, company-register number, legal form, status, founding date, LEI status, and its direct and ultimate parent companies. Costs USD 0.005 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Company Lei
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • company_usUS public company profile from SEC EDGAR by ticker or CIK: industry, exchange, state of incorporation, fiscal year end, address and phone, shares outstanding, latest full-year fina…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
    US public company profile from SEC EDGAR by ticker or CIK: industry, exchange, state of incorporation, fiscal year end, address and phone, shares outstanding, latest full-year financials (revenue, net income, assets, liabilities, equity, cash, diluted EPS) and links to the latest 10-K, 10-Q and 8-K filings. Costs USD 0.01 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Company Us
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • domain_whoisDomain registration data from RDAP (the successor of WHOIS): registrar, creation, expiry and update dates, status codes, nameservers and DNSSEC.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
    Domain registration data from RDAP (the successor of WHOIS): registrar, creation, expiry and update dates, status codes, nameservers and DNSSEC. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Domain Whois
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • embed_textText embeddings with BGE-M3 (multilingual, 1024 dimensions) for search, clustering and RAG: up to 100 texts of up to 8,000 characters per call.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 embeddings with BGE-M3 (multilingual, 1024 dimensions) for search, clustering and RAG: up to 100 texts of up to 8,000 characters per call. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Embed Text
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • extract_dataExtract structured JSON from a web page, PDF or text: name the fields you want (up to 25, with types) and get each value with the exact quote from the page that states it.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
    Extract structured JSON from a web page, PDF or text: name the fields you want (up to 25, with types) and get each value with the exact quote from the page that states it. Every quote is checked against the page itself (verified, quoted, unverified or not_found), so values are never invented. Charged only when at least one field is found. Costs USD 0.005 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Extract Data
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • extract_freeFree.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
    Free. Extract structured JSON from a web page or text with a checked quote per field (up to 8 fields, Qwen3), 20 a day per IP. extract_data is the paid version: gpt-oss-120b, up to 25 fields, no limit.

    Annotations the server declares

    • title: Extract Free
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • file_to_markdownConvert an uploaded document (PDF, Word, Excel, CSV, HTML, XML, OpenDocument; up to 10 MB) to Markdown.Reads text others write1 finding

    Why it is in this group

    • The description's first verb is "convert": a word for a change.
    • 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 "file_to_markdown" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

      Matched text

      annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "convert": a word for a change.

      What this check looks for
    Description, as the server wrote it
    Convert an uploaded document (PDF, Word, Excel, CSV, HTML, XML, OpenDocument; up to 10 MB) to Markdown. Costs USD 0.003 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: File To Markdown
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • fx_ratesCurrency exchange rates and conversion for about 30 currencies (USD, EUR, GBP, INR, JPY, CNY ...) from the European Central Bank reference rates of the latest working day.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
    Currency exchange rates and conversion for about 30 currencies (USD, EUR, GBP, INR, JPY, CNY ...) from the European Central Bank reference rates of the latest working day. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Fx Rates
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • generate_imageGenerate a 1024x1024 JPEG image from a text prompt with FLUX.1 schnell; every prompt passes a safety check first (sexual content and other unsafe requests are refused and not charg…Reads text others write1 finding

    Why it is in this group

    • The name "generate_image" says "generate": a word for a change.
    • 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 "generate_image" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

      Matched text

      annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The name "generate_image" says "generate": a word for a change.

      What this check looks for
    Description, as the server wrote it
    Generate a 1024x1024 JPEG image from a text prompt with FLUX.1 schnell; every prompt passes a safety check first (sexual content and other unsafe requests are refused and not charged). Returns the image as base64. Costs USD 0.01 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Generate Image
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • service_infoFree.No 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.
    Description, as the server wrote it
    Free. Lists every tool with its price and explains how payment works (x402 over MCP, USDC on Base).

    Annotations the server declares

    • title: Service Info
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • transcribe_audioSpeech to text with Whisper large v3 turbo: transcribe audio (mp3, wav, m4a, ogg, webm; up to 20 MB, about 20 minutes) from a URL or base64, or translate the speech into English.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
    Speech to text with Whisper large v3 turbo: transcribe audio (mp3, wav, m4a, ogg, webm; up to 20 MB, about 20 minutes) from a URL or base64, or translate the speech into English. Returns the text, language, duration and timed segments. Costs USD 0.01 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Transcribe Audio
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • translate_textTranslate text (up to 5,000 characters) between 100 languages with M2M100 1.2B.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
    Translate text (up to 5,000 characters) between 100 languages with M2M100 1.2B. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Translate Text
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • verify_emailCheck an email address: format, whether the domain exists and accepts mail (MX records), disposable providers, role accounts and free providers, with a valid / risky / invalid verd…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 an email address: format, whether the domain exists and accepts mail (MX records), disposable providers, role accounts and free providers, with a valid / risky / invalid verdict. The mailbox itself is not probed. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Verify Email
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • watch_fieldWatch one field on a web page for 30 days, e.g.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.

    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
    Watch one field on a web page for 30 days, e.g. a price, stock status or version number: it is read again every 15 to 1440 minutes and an alert goes to your webhook, Slack, Discord or Telegram when it changes, goes below or above a number, contains a text, or disappears. Returns the current value now; check status or stop it for free with the watch secret. Costs USD 0.10 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Watch Field
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
Only reads 7Show 7 read-only tools
  • dns_lookupDNS records of a domain (A, AAAA, MX, TXT, NS, CNAME, SOA, CAA, SRV, PTR) over DNS-over-HTTPS (1.1.1.1), with DNSSEC validation status.Reads text others writeNo findings

    Why it is in this group

    • The name "dns_lookup" says "lookup": 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 "http": it reads web pages, which anyone can write.
    Description, as the server wrote it
    DNS records of a domain (A, AAAA, MX, TXT, NS, CNAME, SOA, CAA, SRV, PTR) over DNS-over-HTTPS (1.1.1.1), with DNSSEC validation status. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Dns Lookup
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • read_urlFree.Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "read_url" says "read": 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 name "read_url" says "read" and "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
    Free. Read a public web page or document (HTML, PDF, Word, Excel, CSV; up to 5 MB) as clean Markdown with its title. Up to 20 reads a minute per IP; for more, url_to_markdown is the paid version with no limit.

    Annotations the server declares

    • title: Read Url
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • screenshotScreenshot of a public web page in a real headless Chrome after it has loaded (JavaScript included): a JPEG of the visible window or the full page, 320 to 1920 pixels wide, as base…Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "screenshot" says "screenshot": 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 "screenshot" 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
    Screenshot of a public web page in a real headless Chrome after it has loaded (JavaScript included): a JPEG of the visible window or the full page, 320 to 1920 pixels wide, as base64. Costs USD 0.003 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Screenshot
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • summarizeSummarize text (up to 60,000 characters) or a public web page or PDF by URL with Llama 3.1 8B, short, medium or long, optionally focused on a topic.Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "summarize" 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.
    • 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 text (up to 60,000 characters) or a public web page or PDF by URL with Llama 3.1 8B, short, medium or long, optionally focused on a topic. Costs USD 0.003 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Summarize
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • url_to_markdownFetch a public web page or document (HTML, PDF, Word, Excel, CSV, XML; up to 10 MB) and return it as clean Markdown with its title and final URL.Reads private dataReads text others writeCan send data outNo findings

    Why it is in this group

    • The description's first verb is "fetch": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "fetch" and "document": it reads documents.
    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • untrusted-inputThe name "url_to_markdown" says "url": it reads web pages, which anyone can write.
    • untrusted-inputThe description's first sentence says "fetch" 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
    Fetch a public web page or document (HTML, PDF, Word, Excel, CSV, XML; up to 10 MB) and return it as clean Markdown with its title and final URL. Add render=true for pages that build their content with JavaScript: they are opened in a real headless Chrome first. Costs USD 0.002 per call, paid with x402 (USDC on Base); calls that fail are not charged.

    Annotations the server declares

    • title: Url To Markdown
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • watch_statusFree.Reads private dataNo findings

    Why it is in this group

    • The name "watch_status" says "status": a word for reading.

    What it can reach

    • private-dataThe input names a secret (input.secret), and the tool reads: it reads secrets.
    Description, as the server wrote it
    Free. A field watch's status, current value and last 20 alerts, with the watch_id and secret that watch_field returned.

    Annotations the server declares

    • title: Watch Status
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false

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.

294 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 2 private-data tools breaks every one.

Reads private data2
Reads files, mail, messages, databases, secrets, calendars or repositories.
Reads text others write21
Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
Can send data out7
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
deny to break every path
{
  "permissions": {
    "deny": [
      "mcp__agent-toolbox__url_to_markdown",
      "mcp__agent-toolbox__watch_status"
    ]
  }
}

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.

Tools listed without sign-in

Looked at Nov 15, 2026, 13:23 UTC, unchanged since Oct 31, 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
17
Protocol version
2025-06-18
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
not observed
Authorization servers
not observed
Legacy HTTP+SSE transport
no
Dynamic client registration
not observed
Client ID metadata documents
not observed
  • HighThe whole server

    The endpoint served tools/list without credentials, and the list has 17 tools that change things (1 destructive, 16 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://agent-toolbox.zafermemon87.workers.dev/mcp: tools/list returned 24 tools to an anonymous client, including stop_watch (destructive), generate_image (write), file_to_markdown (write), transcribe_audio (write), embed_text (write) and 12 more

    What this check looks for

Posture changes

  1. Oct 31, 2026
    • Tools that change things, listed without credentials: 11 to 17

Findings

Matches of fixed checks, with the text they matched.

4 warn
  • The server marks "generate_image" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

    Matched text

    annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The name "generate_image" says "generate": a word for a change.

    What this check looks for
  • The server marks "file_to_markdown" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

    Matched text

    annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "convert": a word for a change.

    What this check looks for
  • The server marks "chat_gpt_oss_120b" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

    Matched text

    annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "open": a word for a change.

    What this check looks for
  • The server marks "chat_qwen3_30b" read-only (readOnlyHint: true), and its own name, description or input give the class "write".

    Matched text

    annotations: {"readOnlyHint":true,"destructiveHint":false,"idempotentHint":false,"openWorldHint":true}; The description's first verb is "open": a word for a change.

    What this check looks for

History

Changes to its tool definitions and to its remote endpoint, newest first.

  1. Tools9 added, 1 changed.+9 added~1 changedView diff
  2. PostureRemote posture changed. Tools that change things, listed without credentials: 11 -> 17.View diff
    • Tools that change things, listed without credentials: 11 to 17

Install

Pinned to version 1.0.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http agent-toolbox https://agent-toolbox.zafermemon87.workers.dev/mcp
claude mcp add --transport http agent-toolbox https://agent-toolbox.zafermemon87.workers.dev/mcp

Run 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.

.claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__agent-toolbox__dns_lookup",
      "mcp__agent-toolbox__company_search",
      "mcp__agent-toolbox__watch_status"
    ],
    "ask": [
      "mcp__agent-toolbox__generate_image",
      "mcp__agent-toolbox__url_to_markdown",
      "mcp__agent-toolbox__file_to_markdown",
      "mcp__agent-toolbox__transcribe_audio",
      "mcp__agent-toolbox__embed_text",
      "mcp__agent-toolbox__translate_text",
      "mcp__agent-toolbox__summarize",
      "mcp__agent-toolbox__verify_email",
      "mcp__agent-toolbox__fx_rates",
      "mcp__agent-toolbox__chat_gpt_oss_120b",
      "mcp__agent-toolbox__chat_qwen3_30b",
      "mcp__agent-toolbox__domain_whois",
      "mcp__agent-toolbox__company_lei",
      "mcp__agent-toolbox__company_us",
      "mcp__agent-toolbox__extract_data",
      "mcp__agent-toolbox__screenshot",
      "mcp__agent-toolbox__watch_field",
      "mcp__agent-toolbox__read_url",
      "mcp__agent-toolbox__extract_free",
      "mcp__agent-toolbox__service_info"
    ],
    "deny": [
      "mcp__agent-toolbox__stop_watch"
    ]
  }
}

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-toolbox. Review it before you use it.

Capability tags behind 22 rules
  • mcp__agent-toolbox__generate_image untrusted-input
  • mcp__agent-toolbox__url_to_markdown private-data, untrusted-input, outbound
  • mcp__agent-toolbox__file_to_markdown untrusted-input
  • mcp__agent-toolbox__transcribe_audio untrusted-input
  • mcp__agent-toolbox__embed_text untrusted-input
  • mcp__agent-toolbox__translate_text untrusted-input
  • mcp__agent-toolbox__summarize untrusted-input, outbound
  • mcp__agent-toolbox__verify_email untrusted-input
  • mcp__agent-toolbox__fx_rates untrusted-input
  • mcp__agent-toolbox__dns_lookup untrusted-input
  • mcp__agent-toolbox__chat_gpt_oss_120b untrusted-input
  • mcp__agent-toolbox__chat_qwen3_30b untrusted-input
  • mcp__agent-toolbox__domain_whois untrusted-input
  • mcp__agent-toolbox__company_search untrusted-input
  • mcp__agent-toolbox__company_lei untrusted-input
  • mcp__agent-toolbox__company_us untrusted-input
  • mcp__agent-toolbox__extract_data untrusted-input, outbound
  • mcp__agent-toolbox__screenshot untrusted-input, outbound
  • mcp__agent-toolbox__watch_field untrusted-input, outbound
  • mcp__agent-toolbox__read_url untrusted-input, outbound
  • mcp__agent-toolbox__extract_free untrusted-input, outbound
  • mcp__agent-toolbox__watch_status private-data

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.zaffy18/agent-toolbox/verify?hash=7e3b0c63a9c84336b4d87831cded3ad26840a707a5a7d075629984db6dbe61a9" | jq -e '.changed == false'

Pinned hash 7e3b0c63a9c8, from Nov 9, 2026, 01:18 UTC.

Watch

Get told when its tools change: a new tool, a new description, a new class.

https://radar.anlyon.com/servers/io.github.zaffy18/agent-toolbox/changes.xml

An Atom feed. Works in any feed reader or Slack's RSS app.

Badge

Anlyon Radar badge for io.github.zaffy18/agent-toolbox
[![Anlyon Radar](https://radar.anlyon.com/badge/io.github.zaffy18/agent-toolbox.svg)](https://radar.anlyon.com/servers/io.github.zaffy18/agent-toolbox)

Source: MCP Registry. Synced Oct 24, 2026, 14:18 UTC.

Agent Toolbox MCP server: questions and answers

Answered from the scan above, not from the publisher's description.

Is the Agent Toolbox MCP server safe?

No scan can promise that. What its definitions show: 24 tools, 17 of which can change, send or delete something, and 4 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 Toolbox MCP server delete data?

Yes. 1 tool can delete, cancel or revoke something: stop_watch.

Can the Agent Toolbox MCP server do things that cannot be undone?

No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.

What tools does the Agent Toolbox MCP server have?

24 tools: 1 destructive, 16 write, 7 read. stop_watch, generate_image, file_to_markdown, transcribe_audio, embed_text, translate_text, verify_email, fx_rates and 16 more.

Does the Agent Toolbox MCP server need sign-in or OAuth?

Its remote endpoint listed its tools to an anonymous client, without sign-in.

Does the Agent Toolbox MCP server send data to outside services?

7 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): url_to_markdown, summarize, extract_data, screenshot, watch_field, read_url and 1 more.

Has the Agent Toolbox MCP server changed its tools?

Yes, once since Anlyon Radar started tracking it, most recently on Nov 9, 2026. Every change is on this page and in its Atom feed.