PentestLLM MCP server
io.github.uplinkswitch/pentestllmuplinkswitchstreamable-httpv0.1.0Scanned
Uncensored models for authorized security testing, prices readable without a key
15 tools: 2 do something that can't be undone, 10 change things and 3 only read.
- Reads private data · 2
- Reads text others write · 6
- Can send data out · 1
- No findings
Tools listed without sign-in. Remote posture
With Anlyon
2 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 PentestLLM MCP server can do
The PentestLLM MCP server is published by the GitHub account uplinkswitch. It has 15 tools. It can do things that cannot be undone, such as sending, paying or publishing (chat and compare) and create or change data (10 tools: media_model, estimate, media_quote, proof and 6 more), but cannot delete data. 3 tools only read.
2 tools read private data (list_models and compare); 6 tools bring in text other people can write (list_models, chat, compare and 3 more); 1 tool can send data out of your system (chat). On its own it has all three parts of a prompt-injection exfiltration path: private data, text others write, and a way out.
None of the fixed checks matched anything in its tool definitions. That is not a safety guarantee.
Its remote endpoint lists its tools without sign-in.
Last scanned Nov 15, 2026, version 0.1.0.
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.
15 tools
Can't be undone 2Claude Code: ask
chatSends messages to any chat model id through POST /v1/chat/completions and returns the answer, the token usage and what the request cost.Reads text others writeCan send data outNo findings
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.
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.
Description, as the server wrote it Sends messages to any chat model id through POST /v1/chat/completions and returns the answer, the token usage and what the request cost. system adds instructions, image_urls lets a model that sees images look at them (a file generate made works), web_search lets the model search while it answers. Billed to the key in the MCP request's Authorization header (Bearer sk-ptl-live-...); without one it fails.
Annotations the server declares
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
compareSends the same conversation (or one prompt) to 2 to 5 chat model ids at once through POST /v1/chat/completions and returns every answer side by side with its token usage, cost and …Reads private dataReads text others writeNo findings
Why it is in this group
- The name "compare" says "compare": a word for reading.
- The description's first verb is "send": a word for sending, and what is sent cannot be taken back.
What it can reach
- private-dataThe description's first sentence says "conversation": it reads messages.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "send" and "conversation": it reads messages, which other people can write.
Description, as the server wrote it Sends the same conversation (or one prompt) to 2 to 5 chat model ids at once through POST /v1/chat/completions and returns every answer side by side with its token usage, cost and time, plus the total. For a second opinion from another model, or to pick the model that answers a kind of question best. Completions are capped at max_tokens each (default 1024). Each model is billed to the key in the MCP request's Authorization header; a model that fails is reported and not charged.
Annotations the server declares
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
Changes things 10Claude Code: ask
embedEmbeds one text or up to 64 through POST /v1/embeddings with an embedding model id from list_models (type embedding) and returns one vector per text, in order, with the token usage…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 Embeds one text or up to 64 through POST /v1/embeddings with an embedding model id from list_models (type embedding) and returns one vector per text, in order, with the token usage and the charge. Billed to the key in the MCP request's Authorization header.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
estimateThe cost of a request on every chat model (or the ids given), cheapest first, with the developer's list cost and the saving.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 The cost of a request on every chat model (or the ids given), cheapest first, with the developer's list cost and the saving. Give input_tokens, or a prompt / messages to count from. With the caller's key, prices are the key's own (price locks included). Live from POST /v1/estimate.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
generateRuns any media model id from list_media_models with its settings (see media_model) through POST /v1/generations, waits up to wait_seconds for the files, and returns the generation:…Reads text others writeNo findings
Why it is in this group
- The name "generate" says "generate": a word for a change.
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 Runs any media model id from list_media_models with its settings (see media_model) through POST /v1/generations, waits up to wait_seconds for the files, and returns the generation: a link per file (kept 30 days), a finished image inline, and the charge. A video that is still running when the wait ends is returned as it stands; call generation with its id to keep waiting. Charged to the key in the MCP request's Authorization header when the model accepts the task; a failed generation is refunded in full. To go from an image to a video or an edit, pass the image's URL in image_urls.
Annotations the server declares
- readOnlyHint: false
- destructiveHint: false
- idempotentHint: false
- openWorldHint: true
generationOne generation by id: its status, progress, files and charge, waiting up to wait_seconds for it to finish first.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 One generation by id: its status, progress, files and charge, waiting up to wait_seconds for it to finish first. Use it after generate answers with a video still running. Needs the key the generation was made with. Live from GET /v1/generations/{id}.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
generationsThe organization's generations, newest first, each with its status, model, files and charge: the asset library behind generate.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 The organization's generations, newest first, each with its status, model, files and charge: the asset library behind generate. Filter by kind, model or status; page with before. Files are kept 30 days. Needs the key. Live from GET /v1/generations.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
media_modelEverything generate needs to call one image, video, speech, music or 3D model: each setting under params with its type, the values it takes, its range and its default, and the pric…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 Everything generate needs to call one image, video, speech, music or 3D model: each setting under params with its type, the values it takes, its range and its default, and the price per unit at each setting under pricing. No key needed. Live from GET /v1/catalog/media.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
media_quoteWhat generate would charge for exactly these settings, at the key's own rates: the units billed, the charge, and the developer's own price for the same request where one is on reco…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 What generate would charge for exactly these settings, at the key's own rates: the units billed, the charge, and the developer's own price for the same request where one is on record. The settings are checked as a real request's would be, so a wrong value fails here, not at the model. Nothing runs and nothing is charged. Needs the key. Live from POST /v1/generations/quote.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
proofWhether each model is still the model it says it is: every day one fixed prompt goes to every model and the verdict per model is stable, changed, failed or unchecked, with the time…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 Whether each model is still the model it says it is: every day one fixed prompt goes to every model and the verdict per model is stable, changed, failed or unchecked, with the time of the last check and how long the answer has been unchanged. No key needed. Live from GET /v1/proof.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
rerankOrders up to 100 documents by relevance to a query through POST /v1/rerank with a rerank model id from list_models (type rerank): each result's index into documents, its score and …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 Orders up to 100 documents by relevance to a query through POST /v1/rerank with a rerank model id from list_models (type rerank): each result's index into documents, its score and its text, best first. Billed per input token to the key in the MCP request's Authorization header.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
transcribeTurns the speech in an audio or video file at an https URL (25 MB at most; a file generate made works) into text through POST /v1/audio/transcriptions, or into English with transla…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 Turns the speech in an audio or video file at an https URL (25 MB at most; a file generate made works) into text through POST /v1/audio/transcriptions, or into English with translate. Use a transcription model id from list_models (type transcription). Billed per second of audio to the key in the MCP request's Authorization header; a request that fails is not charged.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: true
Only reads 3Show 3 read-only tools
list_media_modelsEvery media model on PentestLLM with its id, kind, endpoint, unit and price per image, second, generation or minute at each setting, next to the developer's price where recorded.No findings
Why it is in this group
- The name "list_media_models" says "list": a word for reading.
Description, as the server wrote it Every media model on PentestLLM with its id, kind, endpoint, unit and price per image, second, generation or minute at each setting, next to the developer's price where recorded. No key needed. Live from GET /v1/catalog/media.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
list_modelsEvery chat, embedding, rerank and transcription model on PentestLLM with its id, type, the routes and tools that take it, our price (per million tokens, or per audio minute for tra…Reads private dataReads text others writeNo findings
Why it is in this group
- The name "list_models" says "list": a word for reading.
What it can reach
- private-dataThe description's first sentence says "chat": it reads messages.
- untrusted-inputThe description's first sentence says "chat": it reads messages, which other people can write.
Description, as the server wrote it Every chat, embedding, rerank and transcription model on PentestLLM with its id, type, the routes and tools that take it, our price (per million tokens, or per audio minute for transcription), the developer's list price where recorded, the wire formats it answers (openai, anthropic, gemini) and its specs. Image, video, speech, music and 3D models are in list_media_models. No key needed. Live from GET /v1/catalog.
Annotations the server declares
- readOnlyHint: true
- destructiveHint: false
- idempotentHint: true
- openWorldHint: false
searchOne web search through POST /v1/search: titles, URLs, snippets and publish dates, up to 10 results, optionally only from the last day, week, month or year.Reads text others writeNo findings
Why it is in this group
- The name "search" says "search": 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 "web": it reads web pages, which anyone can write.
Description, as the server wrote it One web search through POST /v1/search: titles, URLs, snippets and publish dates, up to 10 results, optionally only from the last day, week, month or year. Billed as one search to the key in the MCP request's Authorization header (the price in GET /v1/catalog server_tools); a search that fails is not charged.
Annotations the server declares
- 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.
12 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 1 outbound tool breaks every one.
- Reads private data2
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write6
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out1
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
{
"permissions": {
"deny": [
"mcp__pentestllm__chat"
]
}
}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 16, 2026, 14:20 UTC, unchanged since Nov 16, 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
- 12
- 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 12 tools that change things (2 irreversible, 10 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://api.pentestllm.ai/mcp: tools/list returned 15 tools to an anonymous client, including chat (irreversible), compare (irreversible), media_model (write), estimate (write), media_quote (write) and 7 more
Findings
Matches of fixed checks, with the text they matched.
No check matched anything in these tool definitions. That is not a safety guarantee.
History
Changes to its tool definitions and to its remote endpoint, newest first.
No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.
Install
Pinned to version 0.1.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http pentestllm https://api.pentestllm.ai/mcpclaude mcp add --transport http pentestllm https://api.pentestllm.ai/mcpRun 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__pentestllm__list_models",
"mcp__pentestllm__list_media_models",
"mcp__pentestllm__search"
],
"ask": [
"mcp__pentestllm__media_model",
"mcp__pentestllm__estimate",
"mcp__pentestllm__media_quote",
"mcp__pentestllm__proof",
"mcp__pentestllm__chat",
"mcp__pentestllm__compare",
"mcp__pentestllm__embed",
"mcp__pentestllm__rerank",
"mcp__pentestllm__transcribe",
"mcp__pentestllm__generate",
"mcp__pentestllm__generation",
"mcp__pentestllm__generations"
],
"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 pentestllm. Review it before you use it.
Capability tags behind 6 rules
- mcp__pentestllm__list_models private-data, untrusted-input
- mcp__pentestllm__chat untrusted-input, outbound
- mcp__pentestllm__compare private-data, untrusted-input
- mcp__pentestllm__search untrusted-input
- mcp__pentestllm__transcribe untrusted-input
- mcp__pentestllm__generate 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.uplinkswitch/pentestllm/verify?hash=5a7a7d2b8f8778db3f28fa788cbd8cf2990ed749c1e85fe35d1a381bd678210c" | jq -e '.changed == false'Pinned hash 5a7a7d2b8f87, from Nov 15, 2026, 18:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.uplinkswitch/pentestllm/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.uplinkswitch/pentestllm)Source: MCP Registry · repository. Synced Nov 15, 2026, 18:18 UTC.
PentestLLM MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the PentestLLM MCP server safe?
- No scan can promise that. What its definitions show: 15 tools, 12 of which can change, send or delete something, and 0 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 PentestLLM MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the PentestLLM MCP server do things that cannot be undone?
- Yes. 2 tools can send, pay, publish or merge: chat and compare.
What tools does the PentestLLM MCP server have?
- 15 tools: 2 irreversible, 10 write, 3 read. chat, compare, media_model, estimate, media_quote, proof, embed, rerank and 7 more.
Does the PentestLLM MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the PentestLLM MCP server send data to outside services?
- 1 tool can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): chat.
Has the PentestLLM MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 15, 2026.