Context MCP
Local documentation for agents: an MCP server searches a library of doc packages for 100+ libraries and answers offline
- 439 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- The server downloads documentation packages from the community registry and stores them in `~/.context/packages/` while queries stay local
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/neuledge-contextDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The server downloads documentation packages from the community registry and stores them in `~/.context/packages/` while queries stay local
This is third-party code. Review the repository files before installing.
What it does
Context is an MCP server that gives an agent current library documentation instead of a model's outdated knowledge. An example from the project text: a model trained on AI SDK 5 suggests `Experimental_Agent`, while version 6 changed the API to `ToolLoopAgent`. Context finds the right package in a shared registry, downloads it and answers locally. The community registry holds more than a hundred ready documentation packages: Next.js, Nuxt, Astro, React Router, TanStack Query, Prisma, Drizzle, Tailwind CSS, Vitest, Playwright, LangChain and others. Packages are stored as compact SQLite databases in `~/.context/packages/`, and queries run locally and return fast. For your own sources there is the `context add` command: it builds a package from a git repository, a local directory or a site with `llms.txt`, and the finished `.db` file can be handed to a colleague. The server starts with `context serve` and there is an HTTP mode on port 8080 and a Docker image.
Who it is for. Developers working with fast-changing libraries and private internal SDKs who want the agent to check the documentation of the right version.
Good fit when
- The agent writes Next.js middleware and takes code from the right version's docs rather than from model memory
- A company's internal library is private and its documentation is built into a package with `context add ./my-lib`
- You work on a plane or without a stable network: after download packages answer offline
- A team shares a ready documentation package as a `.db` file and a colleague needs no build
Not a fit when
- You need documentation for a single library without a separate server: dropping `llms.txt` into the project is simpler
- The library is not in the registry and has no repository or site to build a package from: building needs a source
- You need live data from an external system: Context answers from prebuilt documentation
Example request
How do I create middleware in Next.js 16? Take the answer from that version's documentation and show a code exampleMCP
- Transport
- stdio, http
- Authentication
- not required
Limitations
The community registry covers popular open source libraries and anything else you build yourself with `context add`. A package has to be downloaded once, after which work goes on without a network. The server is installed globally through npm: `npm install -g @neuledge/context`. Package versions can be chosen by a repository tag and if a build or install fails the previous package stays available. On Windows the system may block replacing an open file, in which case close the reader and retry the install. HTTP mode needs the image from `packages/context/Dockerfile`, built from the monorepo root.
How to disable. Remove the context block from the client config (`claude mcp remove context`) and the package with `npm uninstall -g @neuledge/context`. Downloaded databases live in `~/.context/packages/`.
FAQ
How do I connect Context to Claude Code?
Install the package with `npm install -g @neuledge/context`, then run `claude mcp add context -- context serve`. For Cursor, Windsurf, Zed, Goose and OpenCode the README has its own config blocks.
Do I have to download documentation by hand?
No, for registry packages the agent searches, downloads and queries the right package itself. A manual `context install` is not needed.
How do I add my own library's documentation?
With `context add`: from a git repository (`context add https://github.com/your-company/design-system`), a directory (`context add ./my-project`) or a site with `llms.txt` (`context add https://svelte.dev`).
Where do my queries go?
They do not leave your machine: documents are downloaded once and search runs over local SQLite databases in `~/.context/packages/`.
How do I share a package with the team?
Build the package with the `--save` flag, for example `context add ./my-project --name my-lib --pkg-version 2.0 --save ./packages/`, and pass on the `.db` file; a colleague installs it with `context add ./packages/my-lib@2.0.db`.
Discussion
No questions or reviews yet.
Related
MCP serversUp-to-date, version-specific library docs and code examples in your agent's context, via MCP or a CLI plus skill
An MCP server in one binary: AI reads and edits files across several projects and runs your commands
Docs MCP Server
Grounded Docs MCP Server
A local MCP server that indexes docs from websites, GitHub, npm, PyPI and files and gives the agent version-aware search
An MCP server that turns a large codebase into a semantic feature graph for precise navigation and meaning-based search
A CLI and MCP server that use set rules to pack only the project files a task needs for the model
Library docs from Context7 over REST: about a hundred tokens on demand and no always-on MCP server schemas in context
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Context (this card) | no | not required | 439 | |
| Context7 | yes | API key | 64,735 | Upstash's cloud documentation service where queries go out, rather than local packages that work offline |
| rust-docs-mcp | no | not required | 142 | documentation for Rust crates only rather than packages for hundreds of libraries across languages |
README in short
The text explains the problem of models' outdated knowledge, shows an AI SDK example and describes a quick start: installation and connection to Claude Code, Cursor, Codex, VS Code, Windsurf, Zed, Goose and OpenCode. Next come the community registry, offline mode, building your own packages, Docker and a command reference.