A Rust implementation of Code Context MCP server for semantic code search.

📚 Implementation Details: See IMPLEMENTATION.md for architecture decisions, reserved features explanation, and complete configuration guide.

• 🦀 Pure Rust - High performance, low memory footprint
• 🔍 Semantic Search - Vector-based code search using embeddings
• 🌐 Multi-language - Support for Rust, TypeScript, JavaScript, Python, Go, Java, C++, C#
• 🧠 AST-based Chunking - Intelligent code splitting using tree-sitter
• 📦 MCP Protocol - Compatible with Claude Code, Cursor, and other MCP clients
• 💾 Incremental Indexing - Only re-index changed files using file hashing

┌─────────────────────────────────────────────────────────────┐
│                    Code Context MCP                          │
├─────────────────────────────────────────────────────────────┤
│  MCP Server (JSON-RPC 2.0 over stdio)                        │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐       │
│  │ index_codebase│  │ search_code  │  │ clear_index  │       │
│  └──────────────┘  └──────────────┘  └──────────────┘       │
├─────────────────────────────────────────────────────────────┤
│  Core Components                                             │
│  ┌──────────────────┐  ┌──────────────────┐                 │
│  │ Embedding        │  │ Vector Database  │                 │
│  │ (Ollama/OpenAI)  │  │ (Milvus)         │                 │
│  └──────────────────┘  └──────────────────┘                 │
│  ┌──────────────────┐                                        │
│  │ Code Parser      │                                        │
│  │ (tree-sitter)    │                                        │
│  └──────────────────┘                                        │
└─────────────────────────────────────────────────────────────┘

1. Ollama (for local embeddings)

   # Install Ollama
   curl -fsSL https://ollama.ai/install.sh | sh
   
   # Pull embedding model
   ollama pull nomic-embed-text

2. Milvus (vector database)

   # Using Docker
   docker run -d -p 19530:19530 milvusdb/milvus:v2.3.21

# Install from local source
cd /home/de/works/code-context-mcp
cargo install --path .

# Binary will be installed to ~/.cargo/bin/code-context-mcp
# You can run it directly: code-context-mcp

# Build
cargo build --release

# The binary will be at target/release/code-context-mcp
# Run manually: ./target/release/code-context-mcp

# Run without installing
cargo run --release

Copy .env.example to .env and configure:

cp .env.example .env

Edit .env:

OLLAMA_HOST=http://127.0.0.1:11434
EMBEDDING_MODEL=nomic-embed-text
MILVUS_ADDRESS=http://127.0.0.1:19530
MAX_INDEXED_PROJECTS=10
RUST_LOG=info

Add to ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "code-context-mcp": {
      "type": "local",
      "command": ["code-context-mcp"],
      "enabled": true,
      "environment": {
        "OLLAMA_HOST": "http://127.0.0.1:11434",
        "EMBEDDING_MODEL": "nomic-embed-text",
        "MILVUS_ADDRESS": "http://127.0.0.1:19530",
        "RUST_LOG": "info"
      }
    }
  }
}

Note: MCP uses stdio for communication. The server reads JSON-RPC requests from stdin and writes responses to stdout. Logs are written to stderr and won't interfere with the protocol. No need to specify stdin in the configuration.

claude mcp add code-context-mcp \
  -- npx -y code-context-mcp

Or with custom path:

claude mcp add code-context-mcp-rust \
  -e OLLAMA_HOST=http://127.0.0.1:11434 \
  -e MILVUS_ADDRESS=http://127.0.0.1:19530 \
  -- /path/to/code-context-mcp/target/release/code-context-mcp

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "code-context-mcp": {
      "command": "/path/to/code-context-mcp/target/release/code-context-mcp",
      "env": {
        "OLLAMA_HOST": "http://127.0.0.1:11434",
        "MILVUS_ADDRESS": "http://127.0.0.1:19530"
      }
    }
  }
}

Add to ~/.codex/config.toml:

[mcp_servers.code-context-mcp]
command = "/path/to/code-context-mcp/target/release/code-context-mcp"
env = { OLLAMA_HOST = "http://127.0.0.1:11434", MILVUS_ADDRESS = "http://127.0.0.1:19530" }

Index a codebase directory for semantic search.

Multi-Project Support: Each project is indexed independently. When MAX_INDEXED_PROJECTS is exceeded, the oldest project is automatically evicted (LRU).

{
  "name": "index_codebase",
  "arguments": {
    "path": "/absolute/path/to/codebase",
    "force": false
  }
}

Search the indexed codebase.

Cross-Project Search: Set cross_project: true or use path: "all" to search across all indexed projects.

{
  "name": "search_code",
  "arguments": {
    "path": "/absolute/path/to/codebase",
    "query": "find functions that handle authentication",
    "limit": 10,
    "cross_project": false
  }
}

Clear the search index. Use path: "all" to clear all indexed projects.

{
  "name": "clear_index",
  "arguments": {
    "path": "/absolute/path/to/codebase"
  }
}

Get indexing status. Use path: "all" to see all indexed projects.

{
  "name": "get_indexing_status",
  "arguments": {
    "path": "/absolute/path/to/codebase"
  }
}

This implementation uses a progressive development strategy. Some features are reserved for future use:

• OpenAI Embedding - Currently uses Ollama (local, free). OpenAI provider is implemented but not enabled by default.
• Notification Support - MCP notification protocol is reserved for future push notifications (indexing progress, file changes).
• Roots Capability - Multi-project/monorepo support is planned.
• Variable-level Search - Currently focuses on function/class level. Variable search is reserved.

See IMPLEMENTATION.md for detailed explanations.

# Run in development mode
cargo run

# Run tests
cargo test

# Build release
cargo build --release

# Check code
cargo clippy

# Format code
cargo fmt

MIT