Velvet for AI tools
Your AI assistant, now fluent in Velvet.
Connect Claude, Cursor or VS Code once. Your assistant looks up our real Toy SDK and Velvet Live docs, writes code that works, and adds the Stop button for you.
- ✦ No account or key
- ✦ Read-only
- ✦ Free
// from the Velvet Quickstart toy = await velvet.connect(); await toy.vibrate({ strength: 70, durationMs: 3000 }); stopButton.onclick = () => toy?.stop();
No account, no key
Connect
One URL, every tool.
It's an MCP server: the open standard AI assistants use to call tools. Pick yours and follow the steps.
https://developer.velvetrituals.com/api/mcp- 1Open Add custom connector in Claude (claude.ai or the Claude app: Settings → Connectors).
- 2Name it Velvet and paste the server URL:
https://developer.velvetrituals.com/api/mcp - 3Leave authentication on No sign-in, choose Add, then Connect.
- 4Start a new chat and ask for Velvet code. You’ll see Claude read our docs before it answers.
Try these
Ask for what you’re building.
Each one opens in Claude with the prompt filled in. Claude shows a short safety note for prefilled prompts: that's expected.
A first buzz
“Using the Velvet docs, add a button to my page that connects my Velvet toy and buzzes it at strength 70 for 3 seconds. Load the SDK with a script tag.”
A game moment
“Using the Velvet docs, play the heartbeat pattern from the Velvet cookbook on the player’s toy when they lose a round in my game.”
Sync to video
“Using the Velvet docs, make my Velvet toy follow this <video> element with a toy track.”
Tips to toys
“Using the Velvet docs, wire Velvet Live tips into my PHP cam site: mint the model token on the server and load live.js on her page.”
What it does
Our docs, on call.
It reads. It never controls a toy: that only ever happens in your user's browser, with a Stop button in reach.
search_docsFinds the right section of the Toy SDK or Velvet Live docs.
get_docReads a whole docs page, from Quickstart to Errors.
get_recipesThe cookbook patterns, with the exact code.
get_toy_kindsThe four toy kinds, their models and the command limits.
Every assistant also gets our rules: a working Stop button wherever a toy is controlled, capability checks instead of model names, and secret keys only on your server. Details, troubleshooting and the plain-text docs (llms.txt) are in Use with AI tools. What we log is in our privacy policy.