Code RAG MCP Server

MCP (Model Context Protocol) server for semantic code search. Replace grep/ripgrep with understanding-based code search.

🚀 Features

• ✅ Semantic Search: Find code by concept, not just by text
• ✅ Local Embeddings: Uses LM Studio (no OpenAI required)
• ✅ Multi-language: Go, Python, JS/TS, Terraform, YAML, etc.
• ✅ MCP Integration: Compatible with Claude Code and Zed
• ✅ HTTP API: Git hooks for automatic re-indexing on commit
• ✅ Fast: In-memory indexing with Qdrant

📋 Prerequisites

1. Go 1.22+

   brew install go
   

2. Qdrant (Vector Database)

   docker run -d --name qdrant \
     -p 6333:6333 -p 6334:6334 \
     -v $(pwd)/qdrant_data:/qdrant/storage \
     qdrant/qdrant
   

3. LM Studio with an embedding model

   • Download: https://lmstudio.ai
   • Load model: nomic-ai/nomic-embed-text-v1.5-GGUF
   • Start local server on port 1234

🔧 Installation

# 1. Clone the repo
git clone <your-repo>
cd code-rag-mcp

# 2. Install dependencies
go mod download

# 3. Build
go build -o code-rag-mcp

# 4. Configure
cp config.yaml ~/.config/code-rag-mcp/config.yaml
# Edit config.yaml with your code paths

# 5. Install globally (optional)
sudo cp code-rag-mcp /usr/local/bin/

⚙️ Configuration

1. Config.yaml

Edit ~/.config/code-rag-mcp/config.yaml:

embedding_type: "local"
embedding_model: "nomic-ai/nomic-embed-text-v1.5-GGUF"
embedding_base_url: "http://localhost:1234/v1"
embedding_dim: 768

code_paths:
  - "/Users/you/projects"  # Your code directory

2. Claude Desktop

File: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) File: ~/.config/Claude/claude_desktop_config.json (Linux)

{
  "mcpServers": {
    "code-rag": {
      "command": "/usr/local/bin/code-rag-mcp",
      "env": {
        "LM_STUDIO_URL": "http://localhost:1234/v1"
      }
    }
  }
}

After modification, fully restart Claude Desktop (Quit + relaunch).

3. Claude Code

Option A: Via CLI (Recommended) 🚀

# Add the code-rag MCP server
claude mcp add code-rag \
  --type stdio \
  --command /usr/local/bin/code-rag-mcp \
  --scope user

# Verify installation
claude mcp list
claude mcp get code-rag

Option B: Manual configuration

Claude Code uses the same file as Claude Desktop:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "code-rag": {
      "command": "/usr/local/bin/code-rag-mcp",
      "env": {
        "LM_STUDIO_URL": "http://localhost:1234/v1"
      }
    }
  }
}

Complete copy-paste command (after building):

# 1. Install the binary
sudo cp code-rag-mcp /usr/local/bin/
sudo chmod +x /usr/local/bin/code-rag-mcp

# 2. Add to Claude Code
claude mcp add code-rag \
  --type stdio \
  --command /usr/local/bin/code-rag-mcp \
  --scope user \
  --env LM_STUDIO_URL=http://localhost:1234/v1

# 3. Verify
claude mcp list

4. Zed

File: ~/.config/zed/settings.json

{
  "context_servers": {
    "code-rag": {
      "source": "custom",
      "command": {
        "path": "/usr/local/bin/code-rag-mcp",
        "args": [],
        "env": {
          "LM_STUDIO_URL": "http://localhost:1234/v1"
        }
      }
    }
  }
}

Complete copy-paste command:

# 1. Install the binary (if not already done)
sudo cp code-rag-mcp /usr/local/bin/
sudo chmod +x /usr/local/bin/code-rag-mcp

# 2. Add configuration to Zed
cat << 'EOF' >> ~/.config/zed/settings.json
{
  "context_servers": {
    "code-rag": {
      "source": "custom",
      "command": {
        "path": "/usr/local/bin/code-rag-mcp",
        "args": [],
        "env": {
          "LM_STUDIO_URL": "http://localhost:1234/v1"
        }
      }
    }
  }
}
EOF

# 3. Restart Zed

Note: For Zed, you can also use the "Add Custom Server" button in the Agent Panel.

🎯 Usage

1. Start services

# Terminal 1: Qdrant
docker start qdrant

# Terminal 2: LM Studio
# Open LM Studio UI and start the server

# Terminal 3: Test the server
./code-rag-mcp

2. Use in Claude

Claude: Hi! Let me check the index status.
[Calls: get_index_stats]

You: Index my Terraform project
Claude: [Calls: index_codebase /path/to/terraform]

You: Where is the VPC configuration?
Claude: [Calls: semantic_code_search "VPC network configuration"]

3. Example queries

Semantic search:

"authentication middleware"
"database connection logic"
"error handling patterns"
"terraform AWS VPC modules"
"API rate limiting"

Similar code:

"Find code similar to: func (s *Server) HandleAuth()"

Explanation with context:

"Explain handlers/auth.go with related dependencies"

🔍 Available MCP Tools

semantic_code_search

Primary semantic search. Use instead of grep.

{
  "query": "authentication logic",
  "limit": 5,
  "min_score": 0.7,
  "language": "go"
}

find_similar_code

Find code similar to a given snippet.

{
  "code_snippet": "func HandleError(err error) { ... }",
  "limit": 5
}

explain_code_with_context

Explain a file with its context.

{
  "file_path": "/path/to/file.go",
  "focus": "dependencies"
}

index_codebase

Index a directory. Run this first.

{
  "path": "/Users/you/projects/myapp",
  "extensions": [".go", ".py"]
}

get_index_stats

Check index status.

🧪 Tests

# Test embeddings
go run scripts/test_embeddings.go

# Test indexing
go run scripts/test_indexing.go /path/to/code

# Test searches
go run scripts/test_search.go "authentication"

📊 Recommended Embedding Models

Download in LM Studio

1. Open LM Studio
2. Search: "nomic-embed-text"
3. Download: nomic-ai/nomic-embed-text-v1.5-GGUF (Q8)
4. Load the model in "Local Server"

🐛 Troubleshooting

No results

# Check index
get_index_stats

# Re-index
index_codebase /path/to/code

# Lower minimum score
semantic_code_search "query" min_score=0.5

LM Studio connection fails

# Check if LM Studio is running
curl http://localhost:1234/v1/models

# Check logs
./code-rag-mcp --config config.yaml

Qdrant not responding

# Restart Qdrant
docker restart qdrant

# Check logs
docker logs qdrant

📚 Documentation

• Instructions for Claude
• Integration Guide
• API Reference

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md

📄 License

MIT License - see LICENSE

🙏 Acknowledgments

• MCP Protocol
• Qdrant
• LM Studio
• Nomic AI