Context

Up-to-date docs for AI agents — local, instant, plug and play.

AI agents are trained on outdated docs. When libraries release new versions, your AI doesn't know — and confidently gives you wrong answers.

// Your AI, mass-trained on AI SDK v5 docs, will suggest:
import { Experimental_Agent as Agent, stepCountIs } from 'ai';

// But v6 changed the API entirely:
import { ToolLoopAgent } from 'ai';

The fix isn't better prompting. It's giving your AI the right docs.

How It Works

Context is an MCP server backed by a community-driven package registry with 100+ popular libraries already built and ready to use. When your AI agent needs documentation, it searches the registry, downloads the right package, and queries it locally — all automatically.

Install once. Configure once. Then just ask your AI.

🚀 Quick Start

1. Install

npm install -g @neuledge/context

2. Connect to your AI agent

Context works with any MCP-compatible agent. Pick yours:

Claude Code

claude mcp add context -- context serve

Claude Desktop

Add to your config file:

Linux: ~/.config/claude/claude_desktop_config.json
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "context": {
      "command": "context",
      "args": ["serve"]
    }
  }
}

Restart Claude Desktop to apply changes.

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-specific):

{
  "mcpServers": {
    "context": {
      "command": "context",
      "args": ["serve"]
    }
  }
}

Or use Settings > Developer > Edit Config to add the server through the UI.

OpenAI Codex

Either use the CLI

codex mcp add context -- context serve

Or add to ~/.codex/config.toml (global) or .codex/config.toml (project-specific):

[mcp_servers.context]
command = "context"
args = ["serve"]

Restart OpenAI Codex to apply changes.

VS Code (GitHub Copilot)

Requires VS Code 1.102+ with GitHub Copilot

Add to .vscode/mcp.json in your workspace:

{
  "servers": {
    "context": {
      "type": "stdio",
      "command": "context",
      "args": ["serve"]
    }
  }
}

Click the Start button that appears in the file, then use Agent mode in Copilot Chat.

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

Windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json

{
  "mcpServers": {
    "context": {
      "command": "context",
      "args": ["serve"]
    }
  }
}

Or access via Windsurf Settings > Cascade > MCP Servers.

Zed

Add to your Zed settings.json (press cmd+, or ctrl+, twice):

{
  "context_servers": {
    "context": {
      "command": {
        "path": "context",
        "args": ["serve"]
      }
    }
  }
}

Check the Agent Panel settings to verify the server shows a green indicator.

Goose

Run goose configure and select Command-line Extension, or add directly to ~/.config/goose/config.yaml:

extensions:
  context:
    type: stdio
    command: context
    args:
      - serve
    timeout: 300

OpenCode

Add to ~/.config/opencode/opencode.json:

{
  "mcp": {
    "context": {
      "command": ["context", "serve"],
      "enabled": true,
      "type": "local"
    }
  }
}

3. Ask your AI anything

That's it. Just ask:

"How do I create middleware in Next.js?"

Your agent searches the community registry, downloads the docs, and answers with accurate, version-specific information. Everything happens automatically — no manual context install needed for registry packages.

The Community Registry

The registry is what makes Context plug and play. It's a growing collection of 100+ pre-built documentation packages maintained by the community. Think of it like a package manager, but for AI-ready docs.

Popular packages available today:

Category	Libraries
Frameworks	Next.js, Nuxt, Astro, SvelteKit, Remix, Hono
React ecosystem	React, React Router, TanStack Query, Zustand, Redux Toolkit
Databases & ORMs	Prisma, Drizzle, Mongoose, TypeORM
Styling	Tailwind CSS, shadcn/ui, Styled Components
Testing	Vitest, Playwright, Jest, Testing Library
APIs & Auth	tRPC, GraphQL, NextAuth.js, Passport
AI & LLMs	LangChain, AI SDK, OpenAI, Anthropic SDK

Browse the full registry →

Anyone can contribute. If a library you use isn't listed, submit a PR to add it — your contribution helps every Context user.

Why Local?

Context runs entirely on your machine. Docs are downloaded once and stored as compact SQLite databases in ~/.context/packages/. After that, everything is local.

Fast — Local SQLite queries return in under 10ms
Offline — Works on flights, in coffee shops, anywhere
Private — Your queries never leave your machine
Free — No subscriptions, no rate limits, no usage caps
Reliable — No outages, no API changes, no service shutdowns

Beyond the Registry

The registry covers popular open-source libraries, but Context also works with any documentation source. Use context add to build packages from private repos, internal libraries, websites with llms.txt, or anything not yet in the registry.

# Build from a git repository
context add https://github.com/your-company/design-system

# Build from a local directory
context add ./my-project

# Specific version tag
context add https://github.com/vercel/next.js/tree/v16.0.0

# Build from a website's llms.txt
context add https://svelte.dev

Once built, share packages with your team — they're portable .db files that install instantly:

# Export a package
context add ./my-project --name my-lib --pkg-version 2.0 --save ./packages/

# Teammate installs it (no build step needed)
context add ./packages/my-lib@2.0.db

🐳 Docker

Run Context as a containerized HTTP server for multi-client or Kubernetes deployments.

Pre-built linux/amd64 images are published to GitHub Container Registry by the Docker workflow:

docker run --rm -p 8080:8080 \
  --mount source=context-data,target=/home/node/.context \
  ghcr.io/teejs/context:latest

Tag	Moves when
`latest`, `<version>`	the package version changes on `main`, which is the commit each npm release is published from
`main`, `sha-<short>`	every push to `main` that touches the package

To build the image yourself:

# Run from the repository root (required for the monorepo lockfile)
docker build -t context:local -f packages/context/Dockerfile .
docker run --rm -p 8080:8080 \
  --mount source=context-data,target=/home/node/.context \
  context:local

The image runs as the unprivileged node user. Mount /home/node/.context to persist downloaded documentation packages and server configuration across container replacements. The built-in healthcheck verifies that the HTTP server is accepting requests on port 8080.

For a bind mount instead of the named volume above, make the host directory writable by UID/GID 1000:1000 (for example, with chown 1000:1000) before starting the container.

The container starts Context with HTTP transport on port 8080, accessible at http://localhost:8080/mcp. The image uses a multi-stage build with node:22-bookworm-slim for native module compatibility.

📚 CLI Reference

`context browse <package>`

Search for packages available on the registry server.

# Browse by registry/name
context browse npm/next

# Output:
#   npm/next@15.1.3           3.4 MB  The React Framework for the Web
#   npm/next@15.0.4           3.2 MB  The React Framework for the Web
#   ...
#
#   Found 12 versions. Install with: context install npm/next

# Browse with just a name (defaults to npm)
context browse react

`context install <registry/name> [version]`

Download and install a pre-built package from the registry server.

Registry downloads are limited to 512 MiB by default so a broken or compromised registry cannot fill local storage. Set CONTEXT_MAX_DOWNLOAD_BYTES to a positive byte count to choose a different limit. The declared Content-Length and the bytes actually streamed are both checked. An unset or blank value uses the default.

# Install latest version
context install npm/next

# Install a specific version
context install npm/next 15.0.4

# Install from other registries
context install pip/django

`context add <source>`

Build and install a documentation package from source. Use this for libraries not in the registry, or for private/internal docs. The source type is auto-detected.

From git repository:

Works with GitHub, GitLab, Bitbucket, Codeberg, or any git URL:

# HTTPS URLs
context add https://github.com/vercel/next.js
context add https://gitlab.com/org/repo
context add https://bitbucket.org/org/repo

# Specific tag or branch
context add https://github.com/vercel/next.js/tree/v16.0.0

# SSH URLs
context add git@github.com:user/repo.git
context add ssh://git@github.com/user/repo.git

# Custom options
context add https://github.com/vercel/next.js --path packages/docs --name nextjs

From local directory:

Build a package from documentation in a local folder:

# Auto-detects docs folder (docs/, documentation/, doc/)
context add ./my-project

# Specify docs path explicitly
context add /path/to/repo --path docs

# Custom package name and version
context add ./my-lib --name my-library --pkg-version 1.0.0

Option	Description
`--pkg-version <version>`	Custom version label
`--path <path>`	Path to docs folder in repo/directory
`--name <name>`	Custom package name
`--save <path>`	Save a copy of the package to the specified path

Saving packages for sharing:

# Save to a directory (auto-names as name@version.db)
context add https://github.com/vercel/next.js --save ./packages/

# Save to a specific file
context add ./my-docs --save ./my-package.db

From website (llms.txt):

Many websites publish an llms.txt file with AI-ready documentation. Context auto-detects and fetches it. When the site only provides llms.txt (an index of links rather than the inlined llms-full.txt), Context follows each link and fetches the linked document:

# Auto-fetches llms-full.txt or llms.txt from the site
context add https://svelte.dev
context add https://mui.com/material-ui

# Direct URL to a specific llms.txt file
context add https://svelte.dev/docs/svelte/llms.txt

# Custom package name
context add https://react-aria.adobe.com --name react-aria

From an arbitrary URL (blog posts, articles, raw Markdown):

If no llms.txt is found, Context falls back to fetching the page directly. HTML pages are run through a readability extractor (defuddle) so subscribe CTAs, navigation, and comment widgets don't end up in the package:

# A blog post
context add https://overreacted.io/things-i-dont-know-as-of-2018/

# Raw Markdown from GitHub
context add https://raw.githubusercontent.com/neuledge/context/main/README.md --name context-readme

For subscriber-only content on platforms you have a paid account for, see context auth below.

From URL:

context add https://cdn.example.com/react@18.db

From local file:

context add ./nextjs@15.0.db

Finding the right documentation repository:

Many popular projects keep their documentation in a separate repository from their main codebase. If you see a warning about few sections found, the docs likely live elsewhere:

# Example: React's docs are in a separate repo
context add https://github.com/facebook/react
# ⚠️  Warning: Only 45 sections found...
# The warning includes a Google search link to help find the docs repo

# The actual React docs repository:
context add https://github.com/reactjs/react.dev

Common patterns for documentation repositories:

project-docs (e.g., prisma/docs)
project.dev or project.io (e.g., reactjs/react.dev)
project-website (e.g., expressjs/expressjs.com)

When the CLI detects few documentation sections, it will show a Google search link to help you find the correct repository.

`context list`

Show installed packages.

$ context list

Installed packages:

  nextjs@16.0              4.2 MB    847 sections
  nextjs@15.5              3.9 MB    790 sections
  react@18                 2.1 MB    423 sections

Total: 3 packages (10.2 MB)

Several versions of the same package can be installed side by side; each name is listed newest first. Commands that take a bare name (context query react, context serve --libs react) use the highest installed version.

`context remove <name>`

Remove a package. Pass name@version to remove a specific version. A bare name is only accepted when a single version of that package is installed — otherwise the command lists the installed versions and asks you to pick one, so nothing is deleted by guesswork.

context remove nextjs
context remove nextjs@16.0

`context auth`

Store per-platform cookies or headers so context add <url> can fetch subscriber-only content you have a legitimate account for (e.g., a paid Substack or Medium subscription). Credentials are stored in ~/.context/auth.json with 0600 permissions, and matched by domain (with one level of parent-domain fallback for subdomains).

# Store cookies for a domain
context auth add substack.com --cookies "substack.sid=YOUR_SID"

# Add a custom header too
context auth add medium.com --cookies "sid=..." --header "x-frontend: web"

# List configured auth
context auth list

# Remove auth
context auth remove substack.com

`context serve`

Start the MCP server (used by AI agents).

# Stdio transport (default, for single-client MCP integrations)
context serve

# HTTP transport (for multi-client access over the network)
context serve --http
context serve --http 3000
context serve --http 3000 --host 0.0.0.0

# Restrict the session to a subset of installed packages
context serve --libs react next@15.0.4

Option	Description
`--http [port]`	Start as HTTP server instead of stdio (default port: 8080)
`--host <host>`	Host to bind to (default: 127.0.0.1)
`--libs <names...>`	Restrict the session to a fixed set of installed libraries. Each entry is a name (`react`, exposing every installed version) or `name@version` (`react@18.3.1`, exposing only that version). When set, `search_packages` and `download_package` are hidden so the session is locked to that list. Useful for per-project scoping when you have many packages installed globally.

The HTTP transport uses the MCP Streamable HTTP protocol, enabling multiple clients on the local network to connect to a single server instance. The endpoint is available at http://<host>:<port>/mcp.

The HTTP server bounds its own resource use so clients that disappear without closing their session cannot exhaust it. Tune the limits with environment variables (positive integers; an invalid value is a startup error):

Variable	Default	Description
`CONTEXT_MAX_SESSIONS`	`64`	Maximum concurrent MCP sessions. New sessions beyond this get HTTP 503 with a `Retry-After` header.
`CONTEXT_SESSION_IDLE_TIMEOUT`	`1800`	Seconds a session may sit with no requests and no open SSE stream before the server closes it. A client whose session was closed gets HTTP 404 and must re-initialize.

The HTTP server bounds its own resource use so clients that disappear without closing their session cannot exhaust it. Tune the limits with environment variables (positive integers; an invalid value is a startup error):

Variable	Default	Description
`CONTEXT_MAX_SESSIONS`	`64`	Maximum concurrent MCP sessions. New sessions beyond this get HTTP 503 with a `Retry-After` header.
`CONTEXT_SESSION_IDLE_TIMEOUT`	`1800`	Seconds a session may sit with no requests and no open SSE stream before the server closes it. A client whose session was closed gets HTTP 404 and must re-initialize.

`context query <library> <topic>`

Query documentation directly from the command line. Useful for testing and debugging.

# Query a package (use name@version format from 'context list')
context query 'nextjs@16.0' 'middleware authentication'

# A bare name queries the highest installed version
context query nextjs 'middleware authentication'

# Returns the same JSON format as the MCP get_docs tool

⚙️ Architecture

┌─────────────────────────────────────────────────────────┐
│                     Your Machine                        │
│                                                         │
│  ┌──────────┐    ┌──────────────────┐    ┌────────────┐ │
│  │    AI    │    │   Context MCP    │    │ ~/.context │ │
│  │  Agent   │───▶│     Server       │───▶│  /packages │ │
│  │          │    │                  │    └────────────┘ │
│  └──────────┘    └────────┬─────────┘         │         │
│                           │            ┌──────────┐     │
│                           │            │  SQLite  │     │
│                           │            │   FTS5   │     │
│                           │            └──────────┘     │
└───────────────────────────┼─────────────────────────────┘
                            │ (first use only)
                            ▼
                   ┌────────────────┐
                   │   Community    │
                   │   Registry     │
                   └────────────────┘

First time you ask about a library:

The MCP server searches the community registry
Downloads the pre-built documentation package (a SQLite .db file)
Stores it locally in ~/.context/packages/

Every time after:

FTS5 full-text search finds relevant sections locally
Smart filtering keeps results within token budget
Your AI gets focused, accurate documentation in under 10ms

SQLite engines: Context uses the native better-sqlite3 engine, which reads packages straight from disk. If the native binding cannot be loaded, it falls back to sql.js (WebAssembly), which reads each whole package into memory and is slower. The fallback prints a warning, and context serve reports the engine in use at startup. For deployments that must never run the in-memory engine, set CONTEXT_REQUIRE_NATIVE_SQLITE=1: startup then fails with the load error instead of falling back.

❓ FAQ

Can I use Context with non-JavaScript frameworks like Spring Boot, Django, or Rails?

Yes! Context is language-agnostic. It natively supports Markdown (.md, .mdx), AsciiDoc (.adoc), reStructuredText (.rst), and HTML — no conversion needed.

# Python - FastAPI (Markdown)
context add https://github.com/fastapi/fastapi --path docs/en/docs

# Python - Django (reStructuredText)
context add https://github.com/django/django --path docs

# Java - Spring Boot (AsciiDoc)
context add https://github.com/spring-projects/spring-boot --path spring-boot-project/spring-boot-docs/src/docs

# Rust - The Rust Book
context add https://github.com/rust-lang/book --path src

Point Context at the docs folder with --path and it handles the rest.

Can I contribute package definitions for new ecosystems?

Yes! The registry/ directory has YAML definitions organized by package manager:

registry/npm/ — JavaScript/TypeScript (Next.js, React, Tailwind, etc.)
registry/pip/ — Python (FastAPI, Flask, Django, Pydantic)
registry/maven/ — Java (Spring Boot, JUnit, Micrometer)

To add a package, create a YAML file. Two source types are supported:

Git source — clone a repo at a version tag:

# registry/pip/my-library.yaml
name: my-library
description: "Short description of the library"
repository: https://github.com/org/my-library

versions:
  - min_version: "2.0.0"
    source:
      type: git
      url: https://github.com/org/my-library
      docs_path: docs
    tag_pattern: "v{version}"

ZIP source — download HTML docs from a URL (supports {version} placeholder):

# registry/python/python.yaml
name: python
description: "Python programming language official documentation"

versions:
  - versions: ["3.14", "3.13", "3.12"]
    source:
      type: zip
      url: "https://docs.python.org/3/archives/python-{version}-docs-html.zip"
      docs_path: "python-{version}-docs-html"
      exclude_paths:
        - "whatsnew/**"
        - "changelog.html"

Version discovery is supported for npm, PyPI, and Maven Central. See existing definitions for examples.

🔧 Development

# Install dependencies
pnpm install

# Build
pnpm build

# Test
pnpm test

# Lint
pnpm lint

📄 License

Apache-2.0

Name		Name	Last commit message	Last commit date
.agents/registry		.agents/registry
.changeset		.changeset
.github/workflows		.github/workflows
packages		packages
registry		registry
.clinerules		.clinerules
.cursorrules		.cursorrules
.gitattributes		.gitattributes
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
INSTRUCTIONS.md		INSTRUCTIONS.md
LICENSE		LICENSE
README.md		README.md
SERVER_SPEC.md		SERVER_SPEC.md
biome.json		biome.json
glama.json		glama.json
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
pnpm-workspace.yaml		pnpm-workspace.yaml
pnpm.config.json		pnpm.config.json
server.json		server.json
turbo.json		turbo.json

Navigation Menu

Repository files navigation

Context

How It Works

🚀 Quick Start

1. Install

2. Connect to your AI agent

3. Ask your AI anything

The Community Registry

Why Local?

Beyond the Registry

🐳 Docker

📚 CLI Reference

`context browse <package>`

`context install <registry/name> [version]`

`context add <source>`

`context list`

`context remove <name>`

`context auth`

`context serve`

`context query <library> <topic>`

⚙️ Architecture

❓ FAQ

Can I use Context with non-JavaScript frameworks like Spring Boot, Django, or Rails?

Can I contribute package definitions for new ecosystems?

🔧 Development

📄 License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Context

How It Works

🚀 Quick Start

1. Install

2. Connect to your AI agent

3. Ask your AI anything

The Community Registry

Why Local?

Beyond the Registry

🐳 Docker

📚 CLI Reference

context browse <package>

context install <registry/name> [version]

context add <source>

context list

context remove <name>

context auth

context serve

context query <library> <topic>

⚙️ Architecture

❓ FAQ

Can I use Context with non-JavaScript frameworks like Spring Boot, Django, or Rails?

Can I contribute package definitions for new ecosystems?

🔧 Development

📄 License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

`context browse <package>`

`context install <registry/name> [version]`

`context add <source>`

`context list`

`context remove <name>`

`context auth`

`context serve`

`context query <library> <topic>`