MCP server FAQ

What is the MCP server?

The Model Context Protocol server that provides tools, prompts, and resources to Claude Code and other MCP clients. It handles authentication, telemetry, memory operations, help lookup, and workflow execution.

When do I need to use it?

You use the MCP server whenever you work with Attune AI features through Claude Code. The server runs automatically when you have a .mcp.json file in your project root - you don't need to start it manually.

What tools does it provide?

The server provides several categories of tools:

• Help tools: help_lookup, help_maintain, help_init for contextual documentation
• Memory tools: memory_store, memory_retrieve, memory_search, memory_forget for persistent data
• Utility tools: auth_status, telemetry_stats, attune_get_level for configuration and monitoring
• Workflow tools: Access to all available Attune workflows

How do I start the server manually?

Run uv run python -m attune.mcp.server from your project directory. This is mainly useful for testing - Claude Code normally starts it automatically.

What if the server isn't responding?

Check that your .mcp.json file exists and uses uv run instead of bare python. See the troubleshooting guide for MCP server issues for the complete diagnosis flow.

Where is the server code?

The main implementation is in src/attune/mcp/server.py. Tool definitions are spread across several modules:

• Memory tools: src/attune/mcp/memory.py
• Prompt handling: src/attune/mcp/prompts.py
• Rate limiting: src/attune/mcp/server.py

Tags: mcp, tools, server