MCP server

voicequal ships a small Model Context Protocol server so any MCP-capable agent (Claude Desktop, Claude Code, others) can ask "is this recording clean enough?" as a tool call.

pip install 'voicequal[mcp]'
voicequal-mcp            # runs the stdio server

Tools

Tool	Input	Output
assess_file	path: audio file on the machine running the server	Every FileAssessment field, plus path and advice
assess_base64_wav	data: a WAV file, base64 encoded (a data: URL prefix is fine)	Every FileAssessment field, plus advice

advice is the output of advise(): headline, actions, severity. Both tools raise on a missing file or malformed base64 so the agent sees an error rather than a made-up score.

The base64 tool writes the bytes to a temporary file, assesses it, and deletes it before returning.

Configuring a client

Both Claude Desktop and Claude Code use the same stdio config shape. Point command at the voicequal-mcp executable in the environment where you installed the extra.

Claude Code (.mcp.json in a project, or ~/.claude.json for all projects):

{
  "mcpServers": {
    "voicequal": {
      "command": "voicequal-mcp"
    }
  }
}

Or from the shell:

claude mcp add voicequal -- voicequal-mcp

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "voicequal": {
      "command": "/absolute/path/to/venv/bin/voicequal-mcp"
    }
  }
}

If the server runs inside a Poetry or virtualenv, use the absolute path to that environment's voicequal-mcp (or python -m voicequal.mcp_server) so the desktop app does not pick up a different Python.

Notes

• Importing voicequal never imports mcp; the dependency is only loaded when the server is built.
• Works with mcp 1.x (FastMCP) and 2.x (MCPServer).
• The server is stdio only. No network listener is opened.