fetching-library-docs

Run the fetch-docs.sh script with a library name and topic, for example --library react --topic useState. The script resolves the library ID, fetches the docs, and returns filtered code examples and API signatures.

The full Context7 response stays inside a shell subprocess while awk, grep, and sed extract only code blocks, signatures, and notes. Only the filtered output enters the LLM context, cutting usage from about 934 tokens to about 205 tokens per query.

Any published library indexed by Context7 works, including React, Next.js, Express, Prisma, MongoDB, Vue, Svelte, Fastify, and NestJS. You can pass a name for automatic resolution or a direct ID like /vercel/next.js.

Code mode returns code examples and API signatures and is the default. Info mode returns conceptual explanations and overview paragraphs with fewer examples, which suits learning-oriented questions.

Resolution can fail when the name does not match Context7's index. Try variations such as next.js, nextjs, or next, or pass the exact Context7 library ID directly with --library-id to skip resolution.

Do not use it for exploring a repository's internal source code or local files, since it only fetches published library documentation. For repo internals, use a source exploration tool such as a deepwiki-based research skill instead.