Codegraph is an MCP server designed to parse codebases and construct a local knowledge graph. Developed by Colby McHenry, this free tool is built with TypeScript and includes its own Node.js runtime (Node ≥22.5). It integrates with various AI coding agents like Claude Code and Gemini CLI, offering capabilities for enhanced codebase understanding and interaction. Codegraph stores its graph data in a local SQLite database, providing FTS5 full-text search for efficient querying. It aims to accelerate AI agent workflows and improve developer productivity.

What does Codegraph connect to?

Codegraph functions as an MCP server that integrates with a codebase, parsing its structure to generate a localized knowledge graph. This graph of entities and relationships is stored within a local SQLite database, specifically at ".codegraph/codegraph.db". The database utilizes FTS5 full-text search for efficient data retrieval. Compatible AI clients connect via stdio for local operations or HTTP for remote/shared deployments, with SSE notifications. The server supports a range of AI clients including Claude Code, Cursor, Codex CLI, OpenCode, Hermes Agent, Gemini CLI, Antigravity IDE, and Kiro.

Core features and tools

Codegraph exposes several callable tools to interact with codebases. The index tool builds the knowledge graph of entities and relationships. list_file_entities provides a list of all entities within a specified file, requiring a "path" string. list_entity_relationships details the relationships of a given entity, using "path" and "name" strings. The codegraph_context tool maps specific regions of the codebase. Furthermore, codegraph_explore allows examination of symbols and their interconnections. These tools empower both developers and AI agents to deeply understand code structure.

Who is it for?

Codegraph is primarily intended for AI coding agents such as Claude Code, Cursor, and Gemini CLI, as well as human developers. It enhances agent capabilities by providing structured codebase understanding, reducing exploration rounds, and lowering token consumption. Developers benefit from its ability to assess change impact and navigate large codebases efficiently. The tool's local operation without external API keys makes it suitable for environments prioritizing data privacy. Skip this if your primary need is a cloud-hosted code intelligence solution with managed infrastructure.

How to get started

No credentials are required for Codegraph's core local functionality. For remote HTTP deployments, authentication tokens, such as HARNESS_MCP_AUTH_TOKEN or x-harness-api-key, can be used.

Use cases

• Large codebase navigation: Aids AI agents in quickly locating entry points and call chains, reducing the need for extensive file scanning.
• Change impact assessment: Allows developers and agents to trace callers, callees, and the full impact radius of any symbol before implementing code changes.
• AI agent acceleration and context building: Provides AI agents with a structured understanding of the codebase to reduce exploration rounds and lower token consumption.

Codegraph offers a local, graph-based approach to code intelligence, serving both developers and AI agents. It enhances code understanding, accelerates agent workflows, and helps in impact analysis without relying on external services. The tool's ability to provide precise, relevant code context can significantly improve development efficiency.