| name | context7-mcp |
| description | Current library, framework, and API documentation through the Context7 MCP tools `resolve-library-id` and `query-docs` — signatures, configuration, setup, and code examples from live indexed docs instead of training-data recall. Use for setup or configuration questions, code generation involving a library, API references, or any mention of a specific framework — React, Vue, Next.js, Prisma, Supabase, and peers. General web research beyond indexed library docs belongs to tavily-search. |
[CONTEXT7_MCP]
Library questions resolve against live indexed documentation, never training data: resolve-library-id maps a library name to a Context7 ID, query-docs answers one scoped question against that ID. A known ID in the /org/project or /org/project/version form passes straight to query-docs with no resolution step.
[01]-[FLOW]
- [RESOLVE]:
resolve-library-id takes libraryName plus the full question as query for relevance ranking.
- [SELECT]: Selection ranks by closest name match, higher benchmark score, official over forks, and a version-specific ID when the task names one.
- [QUERY]:
query-docs takes the chosen libraryId and one single-concept question; a multi-concept question splits per concept, capped at three.
- [APPLY]: Fetched signatures and examples land in the answer as verified fact; the library version rides along only when it changes the guidance.
[02]-[QUERY_SHAPE]
query: "auth"
libraryId: /supabase/supabase
query: "How to verify a JWT and read the authenticated user in a server-side route"
Specific, task-shaped queries return targeted sections; bare keywords return noise. Passing the user's own phrasing whole outperforms a paraphrase.
[03]-[FALLBACK]
A library or version absent from the index routes to its own repository or documentation site through tavily-search — never to memory.
