# Rollcage

> Safety for AI agents that hold money on Solana. Live today: the agent wallet check. GET https://www.rollcage.tech/api/wallet-check?address=<wallet> reads a wallet's last 100 mainnet transactions, token accounts and approvals, and returns a score from 0 to 100, a verdict and what a cosigner would have held. Free in beta, no key, read-only. Crash tests, a cosigner and a guarantee are in private beta.

## Live
- [Agent wallet check API](docs.html#wallet-check): GET /api/wallet-check?address=<wallet>; JSON; free in beta, no key; CORS open
- [App](app/index.html): Sign in with Solana, then the wallet check with a list of the wallets this tab checked
- [Landing check](index.html#try): the same check, no sign-in

## For agents
- [Agent quickstart](agents.html): call the check, read the verdict, act on it
- [SKILL.md](agents/SKILL.md): portable skill; its tool table marks live and planned calls
- [OpenAPI](agents/openapi.yaml): the live endpoint under paths; planned ones under x-planned
- [MCP card](agents/mcp.json): planned tools; the MCP server is not published yet. Use the HTTP API today (curl / fetch); the MCP server and skill package ship when the API leaves beta.

## Planned (private beta, not live: these answer 404 today)
- POST /api/crash-tests, GET /api/crash-tests/{id}: adversarial scenarios in a mainnet-clone sandbox
- POST /api/cosign: allow, hold or block a transaction against a mandate
- GET /api/certificates/{id}: public crash-test certificates
- Payment: Beta: calls are free; the x402 flow is the planned paid tier and is not enforced.

## Trust
- [Security](security.html): what runs today, and the cosigner's design
- [Changelog](changelog.html)
- [Pricing](index.html#pricing): planned pricing; the wallet check is free
- [X](https://x.com/rollcage_sol): @rollcage_sol, DMs open for the private beta