The Model Context Protocol server that exposes attune workflows, help, and memory as tools

Failure modes

Symptom	Cause	Fix	Severity
Tools don't appear in Claude Code	The `.mcp.json` entry is missing or the command can't launch	Add/repair the `mcpServers` entry; confirm `python -m attune.mcp.server` runs	high
`RuntimeWarning: coroutine 'EmpathyMCPServer.call_tool' was never awaited`	`call_tool` invoked without `await`	It is a coroutine — `await` it or use `asyncio.run`	high
Tool calls start getting rejected under load	The rate limiter tripped (60 calls / 60 s)	Slow the call rate, or construct with a higher `max_calls`	medium
A tool returns a "path/argument required" error	The tool's own input contract wasn't met	See that tool's feature page; the server just dispatches	medium
Can't tell why the connection failed	Logs aren't on stdout (stdio is the protocol channel)	Read `<tmp>/attune/attune-mcp.log`	low

call_tool is async. Dispatching without await is the common mistake when driving the server from Python.
stdio is the protocol channel. Don't print to stdout from a handler — logs go to the temp log file, not the console.
The server dispatches; tools own their contracts. A tool-level error (bad args) is the tool's, not the server's.

Confirm the server launches: python -m attune.mcp.server.
Confirm registration: the .mcp.json mcpServers entry.
From Python, create_server().tools / get_resource_list() to confirm the surface.
For a connection problem, read <tmp>/attune/attune-mcp.log.
For a single tool failing, consult that tool's feature page.