Current Library Docs for AI Coding — What Context7 Is, and How to Use It.

A documentation service that feeds current library documentation to AI tools. Without current documentation in context, AI tools can rely on stale training knowledge and suggest APIs that have since changed; Context7 narrows that gap by retrieving the documentation for the library — and, when you name one, the version — you specify, at answer time. It ships as an MCP server for editors and agents, a REST API and a TypeScript SDK for server-side use, and a browsable web catalog. The MCP server is open source; the backend that parses and indexes the documentation is a hosted service run by the vendor.

MCP when a person is in the loop: an MCP-capable editor or agent resolves docs interactively as you work. REST API or the TypeScript SDK when software is in the loop: your own service needs documentation retrieval as a feature. The web catalog is for checking coverage and disambiguating same-named entries before you rely on either.

Version pinning is explicit: you name the release in the query — say, the exact version from your lockfile — and the index resolves documentation for it; without a named version you get the library's current documentation. This matters because a version mismatch — an assistant describing a different release than the one you run — is a common cause of wrong AI answers about libraries. Pin the version from your lockfile, not from memory, and the answer describes your reality.

A free API key is available from Context7's dashboard, and paid options are listed on its own pricing page. Limits and plan compositions change, so verify there before building anything on an assumed quota — this page deliberately quotes no prices.

A documentation lookup transmits your query to a third-party service — treat it accordingly. Send the question and the library name, never secrets, credentials, proprietary identifiers or code you cannot share. Context7 publishes its own data-privacy documentation; read it before wiring the service into a workflow that handles sensitive material, and keep the working rule regardless of any vendor's policy.

Two habits. First, resolve the entity before the query: same-named entries exist, so confirm in the catalog that the library id you target is the product you mean. Second, spot-check the citation against the official documentation the entry links — retrieval narrows the version gap, but the primary source remains the authority, and a thirty-second check on the one API you are about to build on is cheap insurance.