blacktop/mcp-tts
Overview
This skill automatically announces plans, resolved issues, and summaries out loud using text-to-speech. It assigns a unique voice per project so you can identify which project is speaking from another room. Providers fallback in order (google, openai, elevenlabs, say) when rate limits or errors occur. Use it proactively after finishing major milestones to keep the team informed without opening the app.
How this skill works
After a major milestone (planning complete, issue resolved, or summary generated) the skill detects the message type and loads a per-project TTS config. It selects the first available provider from the configured provider order, transforms verbose output into conversational speech (~50–100 words), and calls the chosen TTS tool with the configured voice. On failures the skill updates unavailable providers in the config and retries the next provider, persisting auth/config errors so future attempts skip broken providers.
When to use it
- After finalizing a plan or todo list
- When a bug or production issue is resolved
- At the end of a sprint or major task to summarize work
- When you want hands-free, audible notifications during focused work
- During demos or pairing sessions so remote listeners hear status updates
Best practices
- Keep spoken messages short — target ~15–30 seconds (50–100 words)
- Store and review per-project voice assignments so voices remain distinct across projects
- Ensure API keys are configured for primary providers to avoid unnecessary fallbacks
- Prefer the local 'say' fallback for guaranteed playback on macOS without keys
- Tune text transformation to remove URLs, long hashes, and code blocks for natural speech
Example use cases
- Announce a completed implementation plan so teammates hear next steps without opening the project
- Speak a brief summary after resolving a critical bug so stakeholders nearby know it’s fixed
- Read a sprint summary aloud at the end of the day for quick team syncs
- Automatically notify a developer in another room that tests finished and passed
- Provide audible cues during demos when features go from draft to review-ready
FAQ
The skill treats rate limit/quota errors as temporary and moves to the next provider in the order, preserving the original provider in the order but not marking it unavailable unless an auth/config error occurs.
How are voices assigned to new projects?
On first use the skill auto-selects unused voices from the voice pool, records assignments globally, and saves a per-project config with three voices (planning, issue, summary).
Will it speak code or links verbatim?
No. The skill transforms output to be speech-friendly: it removes or replaces URLs, code blocks, long hashes, and file paths with simple, conversational phrases.
