MCP TTS Copilot Summary

Python MCP server for Windows text-to-speech that reads existing Copilot summaries using SAPI voice fallback.

Behavior Summary

• Exposes two MCP tools:
  • speak_text
  • speak_existing_summary
• Consumes already-prepared summary text (no extra summarization pipeline).
• Trigger strategy is MCP-only and instruction-driven.
• Voice fallback policy:
  1. Use Zira if available.
  2. Else use first available SAPI voice.
  3. If no SAPI voice is available, return non-fatal skipped with warning.
• Speech call behavior is non-blocking by default (MCP_TTS_SUMMARY_ASYNC_SPEECH=true), so Copilot can continue responding immediately.

Install

pip

python -m pip install -e .

uv

uv pip install -e .

Run

python -m mcp_tts_copilot_summary.main

Uses stdio transport.

Configuration

Environment variables:

• MCP_TTS_SUMMARY_ENABLED (default: true)
• MCP_TTS_SUMMARY_MIN_PARAGRAPHS (default: 4)
• MCP_TTS_SUMMARY_MIN_NEW_CODE_LINES (default: 100)
• MCP_TTS_SUMMARY_RATE (default: 0)
• MCP_TTS_SUMMARY_VOLUME (default: 100)
• MCP_TTS_SUMMARY_ASYNC_SPEECH (default: true)

Guard criteria used by tools:

• Speak only when paragraph_count > 4 OR new_code_lines > 100.
• If metadata is not provided, tools proceed unless disabled.

VS Code MCP Registration (Example)

See examples/mcp.json for a complete sample. If mcp-tts-copilot-summary is not on PATH, prefer launching with python -m mcp_tts_copilot_summary.main.

For this workspace on Windows, use:

{
  "mcpServers": {
    "tts-copilot-summary": {
      "command": "python",
      "args": ["-m", "mcp_tts_copilot_summary.main"],
      "env": {
        "MCP_TTS_SUMMARY_ENABLED": "true",
        "MCP_TTS_SUMMARY_MIN_PARAGRAPHS": "4",
        "MCP_TTS_SUMMARY_MIN_NEW_CODE_LINES": "100",
        "MCP_TTS_SUMMARY_RATE": "0",
        "MCP_TTS_SUMMARY_VOLUME": "100"
      }
    }
  }
}

This same config is also available at:

• examples/mcp.json
• .vscode/mcp.json

Agent Instruction Snippet

Use this policy in your agent instructions:

Call `speak_existing_summary` (or `speak_text`) only when:
- response has more than 4 paragraphs, OR
- response introduces more than 100 new code lines.

Pass existing summary text directly as `summary_text`.
Do not generate a second summary for TTS.

Also see examples/agent-instruction.md.

Verification

python -m pip install -e .
pytest -q

Publish to PyPI

This package is ready for both pip and uv once published on PyPI.

1. Create a GitHub repository and push this project.
2. In GitHub repository settings, configure trusted publishing for PyPI.
3. The workflow at .github/workflows/publish.yml builds and attempts publish automatically on every push to main (including merges), on release publish, and on manual dispatch.
4. For an actual new upload to PyPI, bump version in pyproject.toml before pushing. If the version already exists, publish is safely skipped.

After publication, installation works globally with either tool:

python -m pip install mcp-tts-copilot-summary
uv pip install mcp-tts-copilot-summary

Manual checks:

1. Start MCP server and run tools/list.
2. Confirm speak_text and speak_existing_summary are listed.
3. Invoke either tool with sample summary text.
4. Confirm Zira is used when available, otherwise first SAPI voice, otherwise skipped warning.

Copy-Ready Smoke Calls

Expected to speak:

{
  "summary_text": "This is a test summary for speech output.",
  "paragraph_count": 5,
  "new_code_lines": 0
}

Expected to skip (threshold not met):

{
  "summary_text": "Short summary.",
  "paragraph_count": 2,
  "new_code_lines": 10
}