The Problem
You know the feeling. Day one on a new codebase. Deadline in two weeks.
You need to find "where authentication happens" but the function is called validateSessionToken(). Grep for "auth" and you get nothing. You spend 20 minutes clicking through files, afraid to touch anything because you don't know what might break.
Traditional code search matches text, not meaning. You have to already know what you're looking for to find it.
The Solution
OpenCodeIntel understands your code the way you do. Search by what code does, not what it's named.
You search: "authentication logic"
It finds: validateSessionToken(), checkJWT(), authMiddleware.ts
You search: "where we handle payments"
It finds: stripe/checkout.ts, processRefund(), PaymentService
You search: "error handling"
It finds: catch blocks, onError(), processFailure()
No regex. No exact matches. Just describe what you're looking for.
Features
Semantic Code Search
Search by meaning, not keywords. Find that function even when you don't know its name.
Dependency Graph
Visualize how your files connect. Understand the architecture at a glance.
Impact Analysis
About to change auth.ts? Know exactly what breaks before you touch it.
Code Style Intelligence
Understand your team's patterns. snake_case or camelCase? How are errors handled? Match conventions instantly.
Works with Claude (MCP)
Connect as an MCP server. Your AI assistant gets real context about your codebase, not just the file you have open.
Research
The retrieval design came out of a week of experiments in December 2025, published with every query set, raw result and the day-by-day log: OpenCodeIntel/retrieval-research.
The headline result is 94% average Hit@1 across 14 open-source codebases (665 queries), using BM25 and embedding fusion with Voyage rerank-2 on top. Most of those queries were generated from function names, so read the caveats in that README before quoting the number. Production search uses a different reranker and is measured by its own regression eval in backend/evals.
Quick Start
Hosted (Fastest)
Go to opencodeintel.com, connect your GitHub, and start searching in under a minute.
Self-Hosted
git clone https://github.com/OpenCodeIntel/opencodeintel.git
cd opencodeintel
cp .env.example .env
# Add your OpenAI + Pinecone API keys
docker compose up -d
Open localhost:3000. Done.
Requirements: Python 3.11+, Node.js 20+, Redis
# Backend
cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python main.py
# Frontend (new terminal) - MUST use Bun!
cd frontend
bun install && bun run dev
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"opencodeintel": {
"command": "python",
"args": ["/path/to/opencodeintel/mcp-server/server.py"],
"env": {
"BACKEND_API_URL": "http://localhost:8000",
"API_KEY": "your-api-key"
}
}
}
}
Why I Built This
I got tired of grep.
Every time I joined a new project or inherited a codebase, I spent hours just figuring out where things lived. The original authors were gone. The docs were outdated. And I was scared to change anything because I couldn't see what depended on what.
I wanted code search that actually understood what code does. Not pattern matching on text. Something that could answer "where do we handle user sessions?" without me already knowing the answer.
OpenCodeIntel is the tool I wish I had. It's open source because I think every developer deserves to understand their code, not just those at companies with dedicated platform teams.
Architecture
Roadmap
Shipped
- Semantic code search with AI summaries
- GitHub OAuth + repo import
- WebGL dependency graph (Sigma.js + ForceAtlas2 + Louvain clustering)
- Dependency Structure Matrix with circular dep detection
- Impact analysis
- Code style intelligence
- Codebase DNA extraction
- Team rules detection (CLAUDE.md, .cursorrules)
- MCP server for Claude Desktop / Cursor
- Real-time indexing progress (WebSocket)
- Full documentation site (12 pages)
- Frontend test suite (Vitest)
Coming Soon
- MCP authentication overhaul (unified config)
- VS Code extension
- Self-hosted Ollama support (no OpenAI required)
- Team workspaces
- GitLab / Bitbucket support
Want to influence the roadmap? Open an issue.
Contributing
Found a bug? Open an issue
Have an idea? Request a feature
Want to contribute code? See CONTRIBUTING.md