Context Engine

Context Engine makes component libraries AI-accessible. AI coding assistants (Claude, Cursor, Copilot) hallucinate props and miss variants when working with custom design systems — because they have no knowledge of your components. Context Engine solves this by extracting accurate component metadata from source code and serving it to AI tools via MCP, so they generate correct code using your actual API.

How It Works

Your Component Code
       ↓
┌─────────────┐   ┌─────────────┐   ┌──────────────┐
│  Extractor  │ → │  Generator  │ → │   Manifest   │
│ react-docgen│   │ Anthropic / │   │   Builder    │
│ + ts-morph  │   │   Gemini    │   │              │
└─────────────┘   └─────────────┘   └──────────────┘
                                            ↓
                               ┌────────────────────────┐
                               │      MCP Server        │
                               │ AI tools query in real │
                               │  time via WebSocket    │
                               └────────────────────────┘

Packages

Package	Description
@context-engine/core	Extraction pipeline, LLM generation, manifest building
@context-engine/db	PostgreSQL + pgvector schema and repositories
@context-engine/server	Hono HTTP API + MCP server

Quickstart

Prerequisites: Node >= 20.19.0, Yarn 1.22.22, Docker

git clone https://github.com/nexus-labs/context-engine
cd context-engine
yarn install
cp .env.example .env   # add your LLM_API_KEY and DATABASE_URL
yarn docker:up         # start postgres + pgvector
yarn db:migrate        # run schema migrations
yarn server:dev        # start the server on :3000

Environment Variables

Variable	Required	Description
DATABASE_URL	Yes	PostgreSQL connection string
LLM_API_KEY	Yes	Anthropic or Google API key for generation
LLM_PROVIDER	No	anthropic (default) or google
PORT	No	Server port (default 3000)

See .env.example for a full example with default values.

Development

yarn build          # Build all packages
yarn dev            # Watch mode
yarn lint           # ESLint
yarn typecheck      # TypeScript check
yarn format         # Prettier
yarn docker:up      # Start postgres
yarn docker:down    # Stop postgres
yarn db:migrate     # Run migrations
yarn server:dev     # Start server in dev mode

Contributing

Sign the CLA, read CONTRIBUTING.md, then open a PR. All contributions are welcome.

Claude Code and OpenCode AI agent configurations are included out of the box — see AI_SETUP.md for the setup guide and slash commands reference.

License

AGPL v3 — free to use and modify, but any service built on top must also be open source under AGPL v3.