Trace MCP is a specialized server tool by Nikolai Vysotskyi designed to augment AI coding agents. Operating on the mcp platform, it functions as a recomputation and reuse layer, delivering a precomputed, framework-aware dependency graph of a codebase. This server runs locally, ensuring data privacy and eliminating the need for default accounts or API keys. It significantly elevates an AI agent's ability to navigate and interact with complex projects, moving beyond basic file-level comprehension. Trace MCP aims to make AI agents more effective in development workflows.

What does Trace MCP connect to?

Trace MCP connects directly to AI coding agents that communicate using the Model Context Protocol (MCP). It operates as an MCP server, serving as a critical recomputation and reuse layer for these AI systems. Its primary function is to provide AI agents with a sophisticated, precomputed, and framework-aware dependency graph of a codebase. This allows agents to gain a deeper understanding of project structure and interdependencies, moving beyond simple file reading to more intelligent code interactions, thus improving the efficiency and accuracy of agent responses.

Core features and tools

Trace MCP exposes 169 MCP tools and 9 resources, dynamically registered based on the frameworks detected within a project. Key code intelligence tools include get_outline, get_symbol, find_usages, and get_change_impact. For decision memory capabilities, it provides mine_sessions, add_decision, query_decisions, and invalidate_decision. The server supports stdio and HTTP, specifically http://127.0.0.1:3741, as transport protocols. It is compatible with AI clients like Claude Code, Cursor, Windsurf, VS Code, Claude Desktop, and any other MCP-compatible agent.

Who is it for?

Trace MCP is for developers and teams who utilize AI coding agents and seek to enhance their agents' understanding of codebases beyond simple file reading. It is particularly useful for those needing AI assistance with complex code navigation, impact analysis, or quality checks across various languages and frameworks. This tool helps agents move beyond simple file-based interactions. Skip this if your AI coding agent does not use the Model Context Protocol.

How to get started

No credentials are required by default. It supports stdio and HTTP transport protocols.

Use cases

• Framework-aware code navigation and impact analysis, enabling agents to understand code structure and predict change impact.
• Safe refactoring, assisting agents in assessing risk, identifying candidates, checking impact, and renaming symbols across files.
• Security and quality gate checks, allowing agents to run checks before code commits or pull requests.

Trace MCP offers a local-first solution for enhancing AI coding agent capabilities by providing a deep, framework-aware understanding of codebases. It operates without requiring accounts or API keys for its core functions, ensuring privacy for source code. By integrating a precomputed dependency graph, it enables agents to perform more sophisticated tasks like impact analysis and safe refactoring. Start by having your AI agent perform a get_outline query on a new project.