Context8

中文介绍

Context8 is an intelligent context platform for private code repositories, similar to Context7. It provides semantic retrieval capabilities for private codebases to AI coding assistants (Cursor, Claude Desktop, Windsurf) via the MCP (Model Context Protocol).

Features

• Private Code Indexing: Connect to local directories or remote repositories, using Tree-sitter for AST-based semantic chunking
• Hybrid Retrieval: Combines vector semantic search with BM25 keyword search, fused via RRF (Reciprocal Rank Fusion)
• MCP Protocol Support: Standard MCP tools that work directly in Cursor/Claude Desktop
• Wiki Generation: Auto-generate wiki documentation for your codebase using LLM
• Web Dashboard: Manage projects, search code, monitor tasks, and configure settings

Screenshots

Home - Project Management

Manage indexed codebases, add local directories or remote repositories.

Context7 like Cookbook

Wiki Documentation

Auto-generated wiki documentation for your codebase, providing structured navigation and comprehensive guides.

Semantic Search

Cross-codebase semantic search with token limit filtering, supporting Code or Info mode.

Task Monitoring

Real-time monitoring of indexing task progress and history.

Settings

Configure AI model providers, including Embedding and LLM API settings.

Project Structure

context8/
├── apps/
│   ├── mcp-server/          # MCP protocol gateway
│   ├── web/                 # Next.js web dashboard
│   ├── worker/              # Background task processor
│   └── cli/                 # Command-line interface
│
├── packages/
│   ├── database/            # Prisma ORM + PostgreSQL
│   ├── embedding/           # Embedding model client
│   ├── indexer/             # Code indexing logic
│   ├── parser/              # Tree-sitter AST parsing + chunking
│   ├── retriever/           # Hybrid retrieval (vector + BM25)
│   ├── types/               # Shared TypeScript types
│   └── vector-store/        # Qdrant vector database client
│
├── mcp/                     # Standalone MCP client package
│
└── deploy/
    └── docker-compose.yml   # Development infrastructure

Quick Start

1. Install Dependencies

# Install pnpm if not installed
npm install -g pnpm

# Install project dependencies
pnpm install

2. Start Infrastructure

# Start PostgreSQL, Redis, Qdrant
pnpm docker:up

3. Initialize Database

# Generate Prisma Client
pnpm db:generate

# Run database migrations
pnpm db:push

4. Start Development Services

# Start all services
pnpm dev

MCP Tools

resolve-library-id

Resolve a library name to a Context8 library ID.

{
  "libraryName": "payment-service"
}

get-library-docs

Fetch documentation for a library by its ID.

{
  "context8CompatibleLibraryID": "/org/project",
  "topic": "authentication",
  "mode": "code"
}

get-wiki-docs

Fetch wiki documentation for a library.

{
  "context8CompatibleLibraryID": "/org/project",
  "topic": "architecture"
}

delete-project

Delete a project and all its indexed data.

{
  "projectId": "project-uuid"
}

IDE Configuration

Cursor / Claude Desktop

Add to your MCP configuration file:

{
  "mcpServers": {
    "context8": {
      "command": "node",
      "args": ["/path/to/context8/apps/mcp-server/dist/bin/local.js"],
      "env": {
        "DATABASE_URL": "postgresql://postgres:postgres@localhost:5432/context8",
        "QDRANT_HOST": "localhost",
        "QDRANT_PORT": "6333"
      }
    }
  }
}

Environment Variables

Copy .env.example to .env and configure:

cp .env.example .env

Main configuration:

| Variable | Description | Default | |----------|-------------|---------| | DATABASE_URL | PostgreSQL connection string | postgresql://postgres:postgres@localhost:5432/context8 | | QDRANT_HOST | Qdrant service host | localhost | | QDRANT_PORT | Qdrant service port | 6333 | | VOYAGE_API_KEY | Voyage AI API Key (code embedding) | - | | OPENAI_API_KEY | OpenAI API Key (for wiki generation) | - |

Tech Stack

• Runtime: Node.js 20+ / TypeScript 5.3+
• Monorepo: pnpm + Turborepo
• Framework: MCP SDK, Next.js (Web)
• Database: PostgreSQL 15 + Prisma
• Vector Database: Qdrant
• Code Parsing: Tree-sitter
• Embedding Model: Voyage code-3
• Task Queue: BullMQ + Redis

Development

# Build all packages
pnpm build

# Type checking
pnpm typecheck

# Run tests
pnpm test

License

MIT