ucm-mcp 0.2.3

Universal Code Mapper - MCP server for AI-agent code navigation

Universal Code Mapper (UCM) MCP Server

UCM is a fast, offline, Universal Code Mapper exposed as a Model Context Protocol (MCP) server. It enables AI coding assistants (like Claude) to navigate, search, and understand your codebases structurally without relying on embeddings or cloud AI dependencies.

✨ Features

• Blazing Fast: Uses tree-sitter for rapid parsing and SQLite FTS5 for full-text search. No embeddings!
• Interactive Web UI: Comes with a built-in network graph visualization to explore your codebase map in the browser.
• Call Graphs & Relationships: Traverse callers, callees, inheritance, and dependencies.
• Framework Aware: Extensively extracts routes and architectural heuristics for Django and React.
• Incremental Indexing: Instant re-indexing of unchanged files using mtime and content hashing.
• Agent-Optimized Tools: Designed specifically for LLMs to easily understand project architecture, route maps, and dead code.

---

📦 Installation

UCM can be installed directly from PyPI. We recommend using uv for the fastest installation and execution:

uv tool install ucm-mcp

Or using standard pip:

pip install ucm-mcp

---

🚀 Quickstart

1. The Interactive Web UI

UCM comes with a rich, interactive web UI to visualize your codebase architecture, view full network graphs, and run architectural analysis.

When you run UCM, the visualizer UI is automatically served in the background!

uvx ucm-mcp
# Or if installed globally:
# ucm-mcp

Upon startup, you will see a banner in your terminal:

============================================================
🚀 UCM Backend is running!
🌐 Open the UI at: https://ucm-ui.netlify.app/?port=8000
============================================================

Click the link to open the Visualizer UI, which will automatically connect to your local backend.

2. Setting up in Claude Desktop (MCP Client)

To give Claude (or any other MCP-compatible AI assistant) access to your codebase architecture, add UCM to your MCP client configuration.

For Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "ucm": {
      "command": "uvx",
      "args": [
        "ucm-mcp"
      ]
    }
  }
}

Note: Even when running inside Claude, the background Web UI will still be active and available to you via the browser!

---

🛠️ Usage & Tools

Once connected, your AI assistant will have access to a powerful suite of tools to understand your code.

First Step:

• ucm_index_project: Indexes the specified directory. This must be called before queries work. (The UI can also trigger this index!)

Exploration Tools:

• ucm_search_symbol / ucm_search_keywords: Powerful search over AST symbols and full-text code.
• ucm_file_map / ucm_directory_map: View the hierarchical layout of the project.
• ucm_get_symbol_info: Retrieve detailed AST information about a specific class, function, or method.

Analysis Tools:

• ucm_find_callers / ucm_find_callees: Trace execution paths and function calls.
• ucm_impact_analysis: Check what breaks when a specific function or class is modified.
• ucm_architecture_summary: Automatically generates a high-level overview of the project structure.
• ucm_route_lookup: Automatically extracts web framework routes (e.g., Django, React).
• ucm_dead_code_detection: Identifies unused functions and classes.

---

⚙️ Advanced Configuration

You can customize the UCM server using the following CLI arguments:

• --port <PORT>: Change the default port (8000) for the UI server.
• --http: Run exclusively in HTTP (SSE) mode instead of standard MCP stdio.
• --data-dir <PATH>: Override the default SQLite storage location (defaults to ~/.ucm).