Skip to content

Context7 MCP: Live Library Docs for Claude Code

By Flowy · Updated 2026-08-28

upstash/context7 is Context7: an MCP server, four overlapping documentation skills, two agents, and a command that fetch a library's current documentation into Claude Code instead of relying on stale training data. Use it when you ask about a specific library's API, or want a dedicated agent to fetch docs without filling up the main conversation.

What is upstash/context7, actually

upstash/context7 is the repository behind Context7, described in its own summary as a platform for up to date code documentation for LLMs and AI code editors (61,207 GitHub stars). Flowy files it under the documentation category, which is exactly what it is: a way to hand an agent a library's current API instead of whatever a model learned during training. Training data has a cutoff date, and real documentation keeps moving after that date, so a model can describe a method that was renamed two releases ago or invent one that never existed. Context7 exists to close that gap by fetching documentation and code examples at the moment they are needed.

The plugin Flowy indexes is 8 components: one MCP server, four skills, two agents, and one command, all different ways of asking the same two questions: which library is this, and what does its documentation say right now.

The MCP server underneath everything

context7 is the MCP server the rest of the plugin sits on top of. The fact pack Flowy indexed carries no one line description for this entry specifically, so this part comes from the repository itself rather than an indexed description: the server matches a library name to the documentation source Context7 tracks for it, then returns the passage or code example relevant to the question asked. Every skill, both agents, and the command below are ways of triggering that same lookup. None of them replace it.

The four skills, and why three barely differ

Three of the four skills, context7-docs, find-docs, and context7-mcp, describe close to the same job in different words: fetch up to date documentation, API references, and code examples the moment a user asks about a specific library, framework, SDK, CLI tool, or cloud service. Their own descriptions even list the same kind of examples, React, Next.js, Prisma, Express, Tailwind, Django, Spring Boot, Vue, and Supabase, as libraries worth checking even though a model already recognizes the name. The fact pack does not explain why the plugin ships three separate skills for one job, so treat this as one capability with three trigger phrases rather than three distinct tools.

context7-cli is the one that actually differs. Its own description covers a command line tool, named ctx7 in the description text, used to fetch library documentation, manage AI coding skills, and configure the Context7 MCP connection itself. It triggers on someone naming "ctx7" or "context7" directly, not only on a library question, which makes it the entry point for setup and configuration rather than a single lookup.

The two agents and the one command

docs-researcher and docs-researcher.agent carry the identical one line description in what Flowy indexed: a lightweight agent for fetching library documentation without cluttering the main conversation. The fact pack does not explain why the plugin registers what reads like the same agent under two names, so treat it as one dispatch target reachable two ways, useful for sending a long documentation fetch off to its own side conversation instead of filling up the thread someone is reading.

The docs command is the direct version of the same idea. Its own description is short: look up documentation for any library. It is a one line entry point into the same MCP server for someone who wants to type a request instead of waiting for a skill to notice a library name in conversation.

When would I actually reach for this

If a library question comes up naturally, the skills fire without any action from you, and which of the three near identical ones fires depends on how its trigger phrase was written, not on anything you choose. If you already know you want documentation and nothing else, the docs command or the docs-researcher agent skip straight to it, and the agent is the better pick when the fetch is likely to run long. Reach for context7-cli specifically when the job is setting up or configuring Context7 itself, or when you are working from a terminal rather than a chat.

How do I install it

Flowy indexes upstash/context7 from its public GitHub repository. It does not own, bundle, or host Context7, and the documentation lookups happen through Context7's own service once the MCP connection is live, not inside Flowy. Install from the listing page linked at the top of this guide, and all 8 components above, the MCP server, all four skills, both agents, and the command, become available in that session.

Common questions

What does the Context7 plugin do in Claude Code?
It adds Context7's documentation lookup to Claude Code: one MCP server plus four skills, two agents, and a command, all fetching current library documentation and code examples instead of relying on a model's training data. The skills fire automatically when a conversation mentions a specific library, framework, SDK, or cloud service. Flowy indexes the plugin from its public `upstash/context7` repository; it does not host or modify it.
What is the difference between context7-docs and find-docs?
Based on their indexed descriptions, there is no real difference: both skills fetch up to date documentation, API references, and code examples the moment a user asks about a specific library, framework, SDK, CLI tool, or cloud service. Both list the same kind of well known libraries as examples worth checking anyway. The plugin does not explain why it ships both under separate names, so it is safer to treat them as one capability with two triggers than as two different tools.
What does the context7-cli skill do?
It is the one skill in this plugin that is not a plain documentation lookup. Its own description covers using a command line tool named ctx7 to fetch library documentation, manage AI coding skills, and configure the Context7 MCP connection itself, triggered by someone naming ctx7 or context7 directly rather than only asking about a library. Reach for it when the job is setup or configuration, not a one off lookup.
Why does upstash/context7 have two docs-researcher agents?
`docs-researcher` and `docs-researcher.agent` carry the exact same one line description in what Flowy indexed: a lightweight agent for fetching library documentation without cluttering the main conversation. The fact pack does not explain why the plugin registers what reads like the same agent under two names, so it is safer to treat this as one agent reachable two ways than as two distinct tools.
How many components does the upstash/context7 plugin add to Claude Code?
8 total: one MCP server (`context7`), four skills (`context7-docs`, `find-docs`, `context7-cli`, and `context7-mcp`), two agents (`docs-researcher` and `docs-researcher.agent`), and one command (`docs`). All of them are different ways of triggering the same underlying documentation lookup.