Find a starting point
Search documentation and working examples by meaning, then read the full reference or source.
Work together
Connect an AI client to a MathPets model. Give it the language reference, a view of the world, and room to experiment alongside you.
Search documentation and working examples by meaning, then read the full reference or source.
Request screenshots at chosen steps, at regular intervals, or as a labeled sheet of thumbnails to compare how a simulation evolves, without changing your current run.
The code window opens when your AI starts reading. Follow its progress and see the changed lines highlight as each edit arrives. Parameter changes and playback actions highlight their controls too.
https://mathpets.world/api/mcp to your AI client as a remote MCP server. In Claude Code, run claude mcp add --transport http mathpets https://mathpets.world/api/mcp, then /mcp and Authenticate. In Claude, add it under Settings → Connectors → Add custom connector.A client that can’t sign in can use the tab’s connection token instead: open Client without sign-in support? in the Connect AI dialog and send it as Authorization: Bearer YOUR_TOKEN. A remote client also needs a publicly reachable MathPets URL; your local server is reachable only from your own computer.
If you edit the code, the AI’s next edit is rejected and it receives your changes. It can reconcile them and explicitly retry. Simply rereading the model cannot skip this check.
AI edits change the draft. Use the editor’s normal Save or Apply button when you want to keep or run it. Previews use the draft, while parameter changes affect the live model.
A signed-in client works in the tab you connected most recently; open Connect AI in another tab and choose Use this tab to move it. It can’t save or publish models or open your saved library. It stays signed in while you use it, for up to 30 days between uses. Sign it out under Signed-in AI clients in any Connect AI dialog.
Connection tokens reach one tab only. They end when you disconnect, leave the page, or after 12 hours; revoke older ones under Other connections.
Semantic search downloads a small embedding model the first time you use it, then runs in your browser. Previews run in a separate worker, support up to 1,000 steps, and stop if they take too long. Models that need external data use the live workspace rather than previews.