Codesearch, developed by Flupkede, is a free MCP server tool built with Rust, designed to provide deep codebase understanding for AI agents. It distributes as self-contained native binaries, embedding hook scripts for various platforms directly within the binary. This server connects to AI clients supporting the Model Context Protocol and can federate with other Codesearch instances. Its primary function is to enable semantic code search across multiple repositories, enhancing agent capabilities for code-related queries.

What does Codesearch connect to?

Codesearch functions as an MCP server, establishing connections with AI clients that support the Model Context Protocol. It is compatible with clients such as OpenCode, Claude Code, and Cursor, as well as any other MCP client. The server can also federate with other running Codesearch instances, allowing for distributed search capabilities. It supports two main transport protocols: stdio, accessed via the "codesearch mcp" command, and Streamable HTTP, available at the "/mcp" endpoint when operating in serve mode.

Core features and tools

Codesearch provides specific callable tools for AI agents. The primary tools are search and get_chunk. The search tool accepts parameters including "query" (string), "mode" ("semantic" | "literal"), "filter_path" (string), "file_glob" (string), and "language" (string), enabling detailed code discovery. The get_chunk tool allows agents to retrieve full code content on demand, supporting in-depth analysis. These tools are central to its function as a multi-repo semantic code search server.

Who is it for?

Codesearch is intended for AI agents, including platforms like OpenCode, Claude Code, and Cursor, and for developers who aim to equip these agents with advanced codebase understanding. It helps agents perform complex code queries and explore multiple repositories effectively. Developers can use it to enhance the code-related capabilities of their AI tools. Skip this if you do not work with AI agents or require deep, multi-repository code understanding for them.

How to get started

The server authenticates using an API key passed as an environment variable, CODESEARCH_SERVE_API_KEY, when binding to a non-localhost address. Access to the MCP streamable-HTTP transport includes DNS-rebinding protection.

Use cases

• Semantic Code Search: Enables agents to perform complex queries like "where is X implemented" or "find usages of Y" across codebases.
• Multi-repo Codebase Exploration: Allows agents to search simultaneously across multiple repositories with results merged for relevance.
• Agent Guidance/Guardrails: PreToolUse hooks steer agents to prioritize Codesearch for code-related queries over generic tools.

Codesearch by Flupkede offers a specialized server for AI agents, focusing on deep code understanding across multiple repositories. Its Rust-based native binaries and MCP protocol support make it a focused solution for enhancing AI agent capabilities. The tool provides semantic search and chunk retrieval, integrating with popular AI clients. Try using the "search" tool with a "semantic" mode query to explore its codebase understanding.