Note: MCP server

Context

Attune AI's Model Context Protocol (MCP) server provides Claude Code with access to workflows, memory, help system, authentication, and telemetry through a structured tool interface. The server implements the MCP specification to expose Attune's capabilities as callable tools within Claude environments.

Architecture

The MCP server uses a mixin-based design around the core EmpathyMCPServer class:

• EmpathyMCPServer — Main server implementation handling MCP protocol, tool routing, and rate limiting
• MemoryHandlersMixin — Provides memory store/retrieve/search/forget tools for cross-session data persistence
• WorkflowHandlersMixin — Exposes workflow execution tools for running Attune AI automation

The server exposes five tool categories:

1. Workflow tools — Execute Attune workflows directly from Claude
2. Utility tools — Authentication status, telemetry stats, session context management
3. Help tools — Progressive documentation lookup, template maintenance, project bootstrapping
4. Memory tools — Persistent storage for patterns, preferences, and cross-agent coordination
5. Rate limiting — Sliding-window limiter preventing API abuse

Integration patterns

Claude Code discovers the server through .mcp.json configuration files in project roots. The server must run as uv run python -m attune.mcp.server to ensure correct package resolution and avoid Python environment conflicts that prevent tool availability.

Rate limiting applies per-tool with a default 60 calls per 60-second window. Voice interfaces skip memory and session tools (defined in _VOICE_SKIP_TOOLS) to avoid interrupting conversational flow.

Source files

• src/attune/mcp/server.py — Core server and entry point
• src/attune/mcp/memory_handlers.py — Memory tool implementations
• src/attune/mcp/workflow_handlers.py — Workflow tool implementations
• src/attune/mcp/prompts.py — MCP prompt definitions
• src/attune/mcp/tool_schemas.py — Tool schema definitions
• src/attune/mcp/rate_limiter.py — Rate limiting implementation