SammySnake-d/fast-context-mcp

What it solves

It enables AI clients (like Claude Code, Cursor, or Claude Desktop) to perform semantic code searches across a local codebase using natural language queries. It bridges the gap between a user's high-level question and the specific files and line ranges where the relevant logic resides, eliminating the need for the user to manually grep or browse directories.

How it works

The project acts as a Model Context Protocol (MCP) server that leverages Windsurf's Devstral API. When a query is received, the server maps the local project directory and generates a directory tree. This context, along with the query, is sent to the Devstral model, which then generates a series of local commands (such as ripgrep, file reads, and glob searches). These commands are executed locally on the user's machine, and the results are fed back to the AI in multiple rounds of refinement until the most relevant file paths and line ranges are identified.

Who it’s for

Developers who use MCP-compatible AI coding assistants and want a more powerful, AI-driven way to navigate and search their local source code without relying on basic keyword matching.

Highlights

• Zero System Dependencies: Bundles ripgrep and tree via npm, ensuring it works across macOS, Windows, and Linux without manual installation.
• Semantic Search: Uses an AI model to iteratively search and refine results rather than simple string matching.
• Automatic Context Management: Automatically adjusts directory tree depth to prevent payload size limits from being exceeded.
• Auto-Discovery of Credentials: Can automatically extract API keys from local Windsurf or Devin installations.