Semantic Code

Provides semantic code search over Python, Rust, and Markdown with incremental indexing and local vector storage.

python

GitHub Stars

python

Language

4 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "vrppaul-semantic-code-mcp": {
      "command": "uvx",
      "args": [
        "semantic-code-mcp"
      ],
      "env": {
        "SEMANTIC_CODE_MCP_DEBUG": "true",
        "SEMANTIC_CODE_MCP_PROFILE": "true",
        "SEMANTIC_CODE_MCP_CACHE_DIR": "<SEMANTIC_CODE_MCP_CACHE_DIR>",
        "SEMANTIC_CODE_MCP_LOCAL_INDEX": "true",
        "SEMANTIC_CODE_MCP_EMBEDDING_MODEL": "all-MiniLM-L6-v2"
      }
    }
  }
}

You can use semantic-code-mcp to index your codebase and perform semantic code search across Python, Rust, and Markdown. It builds embeddings for code chunks, stores them locally, and returns meaning-aware results that blend semantic similarity with keyword and recency considerations.

How to use

You run the MCP server locally and connect to it from your MCP client. The server indexes your project on first search, then serves ranked results that reflect meaning, not just exact text. Use the MCP client to issue search and indexing commands, and integrate Claude Code with your workflow for rapid, semantic code discovery.

How to install

Prerequisites you need before installation: Python is required to run the server locally. You also need the MCP client tooling to start and manage MCP servers.

Install the MCP runtime tooling if you do not have it yet.
Install and run the semantic-code-mcp server using the provided commands for your platform.
Connect Claude Code to the MCP server using the integration steps shown in your client.

Configuration and run notes

The server uses environment variables to customize behavior. Common options include enabling debug logging, choosing the embedding model, and controlling local index storage. You can pass these variables through your MCP client configuration when starting the server.

Tools and endpoints

Core tools exposed by the MCP server include the following actions. Each tool performs a specific step in indexing or searching code.

Troubleshooting and tips

If you encounter indexing delays after the first search, ensure your project path is correct and that new or changed files are detected by the incremental indexing flow. Check that the embedding model loads lazily on the first query and that the vector store initializes properly.

Available tools

search_code

Search code by meaning with auto-indexing on first query. Returns file path, line range, function/m chunk metadata, content, and a score.

index_codebase

Index a codebase for semantic search, reindexing only new/changed files unless force is true.

index_status

Check indexing status for a project and report progress metrics such as is_indexed, files_count, and chunks_count.

Built by

VeilStrat

AI signals for GTM teams