Head to head
agent-device vs ironclaw
Same numbers we show everywhere else, lined up. Ratings come from users, the grade comes from our scanner, adoption comes from public registries.
No reviews yet
MCP server for mobile app automation: verify, control, and debug iOS, Android, TV, and desktop apps
Full listingNo reviews yet
Sandboxed shell exec for MCP clients: run untrusted agent commands in a gVisor container.
Full listingThe short version
Neither has been reviewed yet, so the comparison rests on the scan and the public usage numbers. The safety scan favours ironclaw (70/100 against 55). agent-device has noticeably more public adoption.
| agent-device | ironclaw | |
|---|---|---|
| User rating | No reviews yet | No reviews yet |
| Safety grade | C55/100 | B70/100 |
| Adoption | Established | Growing |
| GitHub stars | 4.7k | 19 |
| Downloads / week | – | – |
| Runs where | Your machine (local) | Your machine (local) |
| Gateway-ready | No | No |
| Tools | — | — |
| Write actions | — | — |
| Price per call | Free | Free |
| Licence | MIT | AGPL-3.0 |
| Last release | 2026-09-19 | 2026-09-18 |
| Category | AI and agents | AI and agents |