CreatorStudio

• 

  The studio state, the conversation, and every MCP tool call with its real latency.

• 

  One voice, five jobs: cameraman, focus puller, sound engineer, teleprompter and director.

• 

  Live-stream safety: 'stop the stream' is refused until the creator confirms.

• 

  The editor package: title, chapters, retakes, protected moments and an AI-editor prompt.

Inspiration

A solo creator is the camera operator, the lighting tech, the sound engineer, the director and the streamer — all at once. The moment you are in front of the lens, your hands are gone. Every studio integration today is a button or a slider, none of them can be touched from the middle of a talking-head shot.

What it does

CreatorStudio is an Alexa+ add-on that turns one spoken sentence into the exact sequence of studio operations: a cameraman and focus puller, a sound engineer, a teleprompter, and a director who calls "cut, take it again" — 18 typed MCP tools. It also drafts a paste-ready title, description, chapters and retakes from what was actually said on the take.

How we built it

A self-hosted MCP server (spec 2025-11-25, Streamable HTTP) in TypeScript on the official SDK: 18 tools with read-only / destructive annotations, one MCP prompt for intent routing, speech in content and structured cards in structuredContent. A pure guard layer refuses destructive actions during a live stream unless the model passes confirm: true. Real OBS control over obs-websocket v5, dispatched with a hard 400 ms timeout. Dry-run by default.

Challenges we ran into

Voice is high-stakes: a misheard "stop" during a live broadcast must never take a studio off air. We separated the decision (a pure, unit-tested guard) from the effect (a swappable driver), and we measured instead of guessing — npm run smoke prints real per-tool latency.

Accomplishments

A complete, testable Alexa+ MCP add-on: 18 tools, 36 tests, sub-millisecond tool calls, and safety behaviour proven by unit and protocol tests.

What's next

Elgato Key Light and Restream drivers behind the same interface, account linking, and batching two spoken steps into one tool call.

Built with

alexa-plus mcp model-context-protocol streamable-http typescript node.js express zod obs-websocket web-speech-api

Built With

• alexa-plus
• express.js
• mcp
• model-context-protocol
• node.js
• obs-websocket
• typescript
• web-speech-api
• zod