Context7

Ask an AI assistant for code against a library that shipped a new major version last month, and it will often hand you functions that no longer exist. The model isn't lying. It's remembering: training data that froze a year ago while the library kept moving.

Context7, built by Upstash, fixes the remembering problem. It's an MCP server (the Model Context Protocol is the standard plumbing that lets AI tools call outside services) with one job: fetch current, version-specific documentation and working code examples straight from the source, and place them in your prompt at the moment the model needs them. Add "use context7" to a request and your assistant looks up the real docs for the version you're actually running before it writes a line. Two tools do the work — one resolves which library you mean, the other pulls its documentation — and setup is a single command, npx ctx7 setup. There's a small CLI for querying docs by hand, too.

It plugs into Cursor, Claude Code, VS Code, and more than thirty other MCP clients, so it slots in beside whatever coding assistant you already use rather than replacing it. The service is free; an optional API key raises the rate limits.

Who is this for? Developers, mostly. You need to be running an AI coding tool that speaks MCP. But it belongs in this catalog because it answers the most common complaint newcomers have about AI-written code: "it made something up." If you're learning to build with an assistant at your elbow, stale documentation is the first wall you'll hit, and this is the door through it.

It pairs naturally with Aider: one tool edits your code, the other makes sure those edits are grounded in documentation that's actually true this year.

The next time your assistant reaches for a deprecated function, it will check first. That's the whole pitch, and it's enough.