claude-context-local

This MCP server provides a local Claude Code integration for semantic code search. It runs the claude-context-local code-search MCP service inside a uv virtual environment, serving as a local backend that Claude Code can talk to via MCP. The server uses multi-language chunking, EmbeddingGemma for embeddings, and FAISS for fast similarity search, all offline on your machine. After installation, you register the MCP server in Claude Code, then index your codebases and query them semantically through Claude Code, with results surfaced from the local embeddings and index. Tools exposed by the MCP server include indexing, incremental updates, and semantic search that interpret natural language queries to locate relevant code chunks across supported languages.

To use it, register the MCP server as code-search in your uv environment, then index your projects using the provided indexing utilities (standalone or via the MCP workflow). Once indexed, you can issue natural language search prompts in Claude Code such as "find all functions related to data serialization in Go and Python" or "show implementations of a specific interface across the repository". The server returns semantically relevant code chunks with metadata like file path, language, and line numbers, enabling precise navigation directly within Claude Code.

Prerequisites:

• Python 3.12+
• curl or wget for the installer script
• Sufficient disk space (1–2 GB for model/cache, plus index data)

Installation steps:

1. Install the MCP server and dependencies via the remote installer:

curl -fsSL https://raw.githubusercontent.com/FarhanAliRaza/claude-context-local/main/scripts/install.sh | bash

If curl is not available, use wget:

wget -qO- https://raw.githubusercontent.com/FarhanAliRaza/claude-context-local/main/scripts/install.sh | bash

2. (Optional) Update an existing installation by running the same installer command again:

curl -fsSL https://raw.githubusercontent.com/FarhanAliRaza/claude-context-local/main/scripts/install.sh | bash

3. After installation, register the MCP server in Claude Code as described in the Quick Start:

claude mcp add code-search --scope user -- uv run --directory ~/.local/share/claude-context-local python mcp_server/server.py

4. Index your codebase using Claude Code and the MCP workflow (no manual commands needed for basic indexing when using Claude Code).

Notes:

• The installer preserves embeddings and indexed projects across updates and can auto-download the EmbeddingGemma model if not cached.
• You can customize the storage location via CODE_SEARCH_STORAGE (see additional notes).

Environment and configuration tips:

• CODE_SEARCH_STORAGE: Set a custom storage directory for embeddings and indexes (default: ~/.claude_code_search). Example: export CODE_SEARCH_STORAGE="/path/to/storage".
• The MCP server runs locally in the uv environment; ensure the directory ~/.local/share/claude-context-local exists and is writable.
• If you plan to use GPU acceleration for FAISS, ensure your system has the required CUDA/driver setup; the installer will attempt to install faiss-gpu when a compatible GPU is detected.
• UF Environment: The MCP integration works offline once the embeddings and index are generated; no external API keys are needed.
• For updates, re-run the installer command; your embeddings and indexed projects are preserved automatically.