Context7 Review

Context7 is a documentation service for AI coding agents, run by Upstash. It indexes the docs and code examples of public libraries and frameworks (Next.js, React, Playwright, PostgreSQL, Cloudflare and many more), and when your agent needs to use one, it fetches the current, version-specific snippets into the agent's context instead of relying on what the model remembers from training. You trigger it by adding "use context7" to a prompt, naming a library ID such as /vercel/next.js, mentioning a version, or setting a rule so the agent calls it automatically.

Setup is one command, npx ctx7 setup, which signs you in, creates an API key and installs either CLI + Skills mode (a skill that tells the agent to run ctx7 commands) or MCP mode (a remote MCP server at mcp.context7.com with resolve-library-id and query-docs tools). It supports Claude Code, Cursor, OpenCode and 30+ other MCP clients, and there is a REST API, a TypeScript SDK and Vercel AI SDK tools. Library maintainers can add or claim their docs, and Pro teams can index private repositories. The MCP server code is MIT-licensed on GitHub; the API backend, parser and crawler are private.

Search API and Context API requests count the same, one call each. Allowances are per seat and not pooled: if one Pro member makes 3,000 calls and another none, 1,000 calls are billed ($5). Context7 says Enterprise starts around $30 per user per month for small teamspaces and can go as low as $2.50 per user for large organisations, with up to 5,000 members per teamspace; teamspaces up to 50 members can activate Enterprise self-serve. You can cancel Pro any time; unpaid usage is charged before the move to Free.

Prices checked on the vendor's pricing page on November 10, 2026. See Context7 plans.

What Context7's own pages and README list:

Context7 suits developers who use AI coding agents with fast-moving frameworks, where models trained months ago suggest outdated or invented APIs. The free tier's 1,000 calls a month covers light individual use; teams that need private repos or never want to be blocked pay $10 per seat. Because indexed docs are community-contributed, treat anything your agent pulls from a little-known library with the same care as other third-party content.

If you want to ask questions about a specific GitHub repository rather than look up library docs, DeepWiki and GitMCP cover that. Ref is another documentation MCP server. If you code in Cursor, its built-in docs indexing may be enough for a few libraries.

Context7's open-source MCP server has about 64,500 stars on GitHub, a sign of wide adoption among developers using coding agents. Its main public incident was security: in February 2026 Noma Security reported "ContextCrush", a flaw where poisoned custom rules on a library could instruct agents to read .env files and delete local files. Noma says Upstash fixed it in production on February 23, 2026, five days after the report, with no exploitation observed; it is tracked as CVE-2026-75130.