Context7
by Context7
Up-to-date library documentation for coding agents. Two read-only tools — resolve a library ID, then query its docs. The directory says sign-in required; the published endpoint answered anonymously on 2026-08-18.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Context7 via MCP
https://mcp.context7.com/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Context7 Tools & Capabilities (2)
Resolves a package/product name to a Context7-compatible library ID and returns matching libraries. You MUST call this function before 'Query Documentation' tool to obtain a valid Context7-compatible library ID UNLESS the user explicitly provides a library ID in the format '/org/project' or '/org/project/version' in their query. Each result includes: - Library ID: Context7-compatible identifier (format: /org/project) - Name: Library or package name - Description: Short summary - Code Snippets: Number of available code examples - Source Reputation: Authority indicator (High, Medium, Low, or Unknown) - Benchmark Score: Quality indicator (100 is the highest score) - Versions: List of versions if available. Use one of those versions if the user provides a version in their query. The format of the version is /org/project/version. For best results, select libraries based on name match, source reputation, snippet coverage, benchmark score, and relevance to your use case. Selection Process: 1. Analyze the query to understand what library/package the user is looking for 2. Return the most relevant match based on: - Name similarity to the query (exact matches prioritized) - Description relevance to the query's intent - Documentation coverage (prioritize libraries with higher Code Snippet counts) - Source reputation (consider libraries with High or Medium reputation more authoritative) - Benchmark Score: Quality indicator (100 is the highest score) Response Format: - Return the selected library ID in a clearly marked section - Provide a brief explanation for why this library was chosen - If multiple good matches exist, acknowledge this but proceed with the most relevant one - If no good matches exist, clearly state this and suggest query refinements For ambiguous queries, request clarification before proceeding with a best-guess match. IMPORTANT: Do not call this tool more than 3 times per question. If you cannot find what you need after 3 calls, use the best result you have.
Retrieves and queries up-to-date documentation and code examples from Context7 for any programming library or framework. You must call 'Resolve Context7 Library ID' tool first to obtain the exact Context7-compatible library ID required to use this tool, UNLESS the user explicitly provides a library ID in the format '/org/project' or '/org/project/version' in their query. Do not call this tool more than 3 times per question.
Read from the server on 2026-08-18, including each tool's own safety annotations.
Limits
- It retrieves documentation; it reads none of your code. Neither tool accepts a file, a repository path or a credential. Context7's data privacy page states that it does not store source files and that your prompts, code and conversation history are never transmitted — only the query the client formulates, the library name or ID, and client metadata.
- Context7 disclaims the accuracy of what it returns. Its repository disclaimer states that projects are community-contributed, that it cannot guarantee accuracy, completeness or security, and that projects are maintained by their owners. It asks users to report suspicious content through a Report button on the project page. Carried here unsoftened, because an agent writing code from these snippets inherits the caveat.
- The free plan caps at 1,000 API calls a month. Context7's pricing page states free users are blocked on reaching that limit, with 20 bonus calls granted daily while blocked until the month resets. Pro is $10 per seat per month with 5,000 included calls per seat and $10 per additional 1,000. Private repository parsing is billed separately at $5 per 1M tokens.
- Anonymous callers get the lowest rate limit and no private sources. Context7's API guide describes the anonymous tier as low limits with no custom configuration. Private repositories require a Pro or Enterprise plan; they are not reachable without a key at all.
- Prompts and resources are declared but empty.
initializedeclarespromptsandresourceswithlistChanged: true;prompts/list,resources/listandresources/templates/listall returned empty arrays on 2026-08-18. An undeclared capability returns error-32601, and none of these did. A prompt picker for this server will be blank. - The published endpoint's
WWW-Authenticateheader is misleading to strict clients. It appears on a successful HTTP 200. A client treating it as a gate will demand a sign-in the endpoint does not require. - Documentation freshness is a threshold, not a guarantee. Context7's refresh runs in the background on request, at one day for top-100 libraries and up to 45 days for everything outside the top 5,000; website-sourced libraries use a slightly higher threshold still. A niche library's documentation can be over a month old and the tools give no signal of its age.
- Private libraries are never refreshed automatically. Context7 states this explicitly; updating one requires a manual refresh from its library page.
- Your queries are stored and passed to third-party LLMs. Context7's data privacy page says client-formulated queries are sent to OpenAI, Google Gemini and Anthropic for reranking, and are anonymously stored for benchmarking. Disabling that storage is an Enterprise-plan control, and Context7 warns it may degrade retrieval quality.
- Anthropic's directory records this connector as requiring authentication, which contradicts what its own published endpoint does. Documented above with both observations; we do not know which will change.
- We did not call either tool. Our check was a read-only
initialize,tools/list,prompts/list,resources/listandresources/templates/listhandshake, plus aGET /ping. Every behavioural claim above comes from the server's own schemas and descriptions, or from Context7's published documentation.
Frequently asked questions
Does the Context7 MCP server require an API key?
No, not at the endpoint Anthropic publishes. Our anonymous handshake against https://mcp.context7.com/mcp on 2026-08-18 returned both tool definitions with no credential attached, and Context7's own client documentation says the standard endpoint works without a key at the anonymous rate limit. A key raises those limits and unlocks private repositories, but it is not needed to connect.
Why does Anthropic's directory say Context7 needs authentication?
Because Context7 runs two endpoints and the posture differs between them. The directory records auth_required for the connector, yet the URL it publishes answered our anonymous handshake with HTTP 200 on 2026-08-18. A second endpoint, the same host with /mcp/oauth, returned HTTP 401 and JSON-RPC error -32001 to the identical anonymous request. Both facts are real.
Why must you call resolve-library-id before query-docs?
Because query-docs takes a Context7 library ID that nothing else on the surface produces. Its required libraryId parameter expects the form /org/project, such as /vercel/next.js, and resolve-library-id is the tool that turns a package name into one. Context7 documents one exception: if you already know the ID, you can pass it straight to query-docs and skip resolution entirely.
What does Context7 send to its servers when my agent uses it?
Only the fields in the two tool schemas plus client metadata. Context7's data privacy page lists the query, the library name or ID, your API key if you supplied one, your MCP client name and version, the transport type, and an AES-256-CBC encrypted client IP used for rate limiting. It states your full prompts, source code and conversation history are never transmitted.
Is the documentation Context7 returns guaranteed to be accurate?
No, and Context7 says so directly. Its repository disclaimer states that projects are community-contributed, that it cannot guarantee the accuracy, completeness or security of all library documentation, and that projects are maintained by their owners rather than by Context7. It asks users to report suspicious content and says you use the service at your own discretion and risk.
How many free requests does the Context7 free plan include?
One thousand API calls a month. Context7's pricing page states that free users are blocked on reaching that monthly limit, and that twenty bonus calls are granted each day while blocked until the month resets. Pro is ten dollars per seat per month with five thousand included calls per seat, then ten dollars per additional thousand.
Are the Context7 MCP tools safe to auto-approve?
Both carry the full four-annotation set read live on 2026-08-18: readOnlyHint true, destructiveHint false, idempotentHint true and openWorldHint true. Neither writes anything and neither accepts a credential in its schema. The residual risks are context poisoning from retrieved third-party text and quota spend, not damage to anything you own.
Sources
- Context7, "Intro" (retrieved 2026-08-18, HTTP 200; Markdown variant at
overview.md). This is the exactdocumentationvalue in Anthropic's directory entry, and it resolves without redirect. Source for the product framing and theuse context7prompt convention. · retrieved 2026-08-18 - Context7, "MCP Clients" (retrieved 2026-08-18, HTTP 200). Per-client install commands, the
Authorization: Bearerheader form, the/mcpversus/mcp/oauthendpoint choice, and the statement that the standard endpoint works without an API key at the anonymous rate limit. · retrieved 2026-08-18 - Context7, "Set Up OAuth" (retrieved 2026-08-18, HTTP 200). The documented endpoint swap to
/mcp/oauth, the stdio exclusion, and the warning that most clients need explicit authentication after setup. · retrieved 2026-08-18 - Context7, "API Guide" (retrieved 2026-08-18, HTTP 200). Library ID formats including websites,
llms.txt, npm and uploaded sources; the two version-pin syntaxes; the anonymous-versus-keyed rate-limit tiers; and the429response headers includingRetry-After. · retrieved 2026-08-18 - Context7, "Best Practices" (retrieved 2026-08-18, HTTP 200). The client rule that avoids typing
use context7, the library-ID shortcut, and version matching. · retrieved 2026-08-18 - Context7, "Keeping Libraries Fresh" (retrieved 2026-08-18, HTTP 200). The refresh thresholds by popularity rank, the background-refresh behaviour, and the statement that private libraries are not refreshed automatically. · retrieved 2026-08-18
- Context7, "Troubleshooting" (retrieved 2026-08-18, HTTP 200). The
/pinghealth check, the401guidance namingAuthorization: Bearerand thectx7skkey prefix, and the recommendation to prefer the remote endpoint over local npx. · retrieved 2026-08-18 - Context7, "Data Privacy" (retrieved 2026-08-18, HTTP 200). The enumerated list of what is sent to the API, the AES-256-CBC encrypted client IP, the reranking providers, the query-storage-for-benchmarking statement, and the no-source-file storage claim. · retrieved 2026-08-18
- Context7, "Security" and "Data Safety" — and (retrieved 2026-08-18, HTTP 200). Upstash SOC 2 infrastructure, and the layered prompt-injection and malware classifier applied to indexed documentation. · retrieved 2026-08-18
- Context7, "Claude Code" (retrieved 2026-08-18, HTTP 200). The plugin contents — skill,
docs-researcheragent,/context7:docscommand — thecontext7-marketplaceinstall route, and the note that the plugin works anonymously when no key is exported. · retrieved 2026-08-18 - Context7, "Pricing & Plans" (retrieved 2026-08-18, HTTP 200; served as Markdown with
Accept: text/markdown, and the target of a 307 fromhttps://context7.com/docs/plans-pricing.md). The sole source for every price and quota on this page: the free tier's 1,000 monthly API calls and 20 daily bonus calls while blocked, Pro at $10 per seat with 5,000 included calls, $10 per additional 1,000, and private-repo parsing at $5 per 1M tokens. · retrieved 2026-08-18 upstash/context7repository (README retrieved 2026-08-18, HTTP 200; repository metadata via the GitHub API confirms an MIT licence and an active push the same day). Source for the two tool names with their required parameters, and for the accuracy disclaimer quoted in Limits. Note the README's own scope statement: the repository hosts the MCP server's source only, while the API backend, parsing engine and crawling engine are private. · retrieved 2026-08-18- Context7 machine affordances —
https://context7.com/llms.txt,https://context7.com/docs/llms.txtandhttps://context7.com/.well-known/integrations.json(all retrieved 2026-08-18, HTTP 200). The docs index drove every fetch above.integrations.jsondeclares the MCP surface's authstatusasrequiredwith aCONTEXT7_API_KEYheader, which is the machine-readable half of the header disagreement noted on this page. · retrieved 2026-08-18 - Live handshake against
https://mcp.context7.com/mcp— anonymousinitializereturned HTTP 200, protocol2025-06-18, serverContext7 4.0.1, capabilitiesprompts,resourcesandtoolseach withlistChanged: true, plus an instructions block;tools/listreturned two tools with full descriptions, input schemas and four-hint annotations;prompts/list,resources/listandresources/templates/listeach returned an empty array (2026-08-18). The 200 response carriedWWW-Authenticate: Bearer resource_metadata=....GET https://mcp.context7.com/pingreturned{"status":"ok","message":"pong"}. Tool schemas committed atdocs/marketing/data/tool-schemas/context7.tools.json. · retrieved 2025-06-18 - Auth probe — the identical anonymous
initializeagainsthttps://mcp.context7.com/mcp/oauthreturned HTTP 401 with JSON-RPC error-32001"Authentication required" (2026-08-18).https://mcp.context7.com/.well-known/oauth-protected-resourcereturned HTTP 200 naming authorization serverhttps://context7.comand scopesprofileandemail;https://context7.com/.well-known/oauth-authorization-serverreturned HTTP 200 with the RFC 8414 metadata. Under this catalogue's rule an RFC 9728 descriptor is the server's own statement, so the OAuth path is verified — and so, separately, is the published endpoint's anonymous success. · retrieved 2026-08-18 - Upstash privacy policy — , the
privacy_policyvalue in Anthropic's directory entry, and the Context7 Addendum at . Both returned HTTP 200 as PDFs on 2026-08-18 and were confirmed reachable but not parsed; we make no claim about their contents. · retrieved 2026-08-18 - Anthropic Connectors Directory entry — . Tier, categories, endpoint, transport, auth posture,
prompt_names: []and both tool names, read from our committed directory snapshot dated 2026-08-16. It records the connector as added 2026-02-06. · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Developer Tools
- Developer
- Context7
- Tools
- 2
- Domain
- mcp.context7.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.