@phoenixaihub/graphrag

Graph RAG for private codebases — index your code into a knowledge graph with semantic search and an MCP server. Self-hosted, no code leaves your machine.

Features

• 🔍 Index TypeScript, JavaScript, and Python codebases into a knowledge graph
• 🧠 Semantic search over code with graph-aware context expansion
• 🔗 Graph relationships — functions, classes, imports, exports, call chains
• 🤖 MCP server — plug into Claude, Cursor, or any MCP-compatible client
• 🏠 Self-hosted — SQLite-backed, runs locally, zero cloud dependencies
• ⚡ Optional embeddings via OpenAI text-embedding-3-small

Quick Start

npm install -g @phoenixaihub/graphrag

# Index a repo
graphrag index ./my-project

# Search code
graphrag query "authentication middleware"

# Start MCP server
graphrag serve --db ./my-project/.graphrag/index.db

MCP Configuration

Add to your Claude Desktop / Cursor config:

{
  "mcpServers": {
    "graphrag": {
      "command": "graphrag",
      "args": ["serve", "--db", "/path/to/project/.graphrag/index.db"]
    }
  }
}

Available MCP Tools

| Tool | Description | |------|-------------| | search_codebase | Semantic search across indexed code entities | | get_function_context | Get a function/class with full graph context (callers, callees, imports) | | find_usages | Find all usages of a function, class, or variable |

With Embeddings

Set OPENAI_API_KEY and use --embed:

export OPENAI_API_KEY=sk-...
graphrag index ./my-project --embed

How It Works

1. Parse — Extracts functions, classes, imports/exports using regex-based parsing
2. Graph — Builds entity relationship graph (contains, imports, calls, extends)
3. Store — Persists to SQLite with full-text search indexes
4. Search — Text search + graph traversal for context-rich results
5. Serve — MCP server exposes tools for AI-assisted code exploration

Comparison

| Feature | GraphRAG | Sourcegraph | Codeium | |---------|----------|-------------|---------| | Self-hosted | ✅ Local SQLite | ❌ Server required | ❌ Cloud | | Privacy | ✅ No code leaves machine | ⚠️ Depends on deployment | ❌ Cloud-processed | | MCP support | ✅ Built-in | ❌ | ❌ | | Setup time | < 1 min | Hours | Minutes | | Cost | Free | $$$$ | Free tier limited | | Graph relationships | ✅ | ✅ | ❌ | | Languages | TS/JS/Python | 30+ | 70+ |

Programmatic API

import { GraphDB, parseFile } from '@phoenixaihub/graphrag';

const db = new GraphDB('.graphrag/index.db');
const result = parseFile('./src/main.ts');

for (const entity of result.entities) {
  db.insertEntity(entity);
}

const results = db.searchEntities('handler', 5);
console.log(results);

License

MIT