agent-device vs Stagehand
Both are Grade A and 2.7 points apart, so choose on what you weigh most. agent-device leads on supply-chain integrity and transparency; Stagehand leads on adoption and rests on broader evidence. Note: agent-device's evidence coverage is thinner (coverage B vs A) — its score rests on fewer independent signal types.
Mobile app automation and verification for AI coding agents. CLI, MCP server, and typed Node.js API for iOS, Android, HarmonyOS, TV, web, macOS, and Linux.
Choose agent-device if supply-chain integrity and transparency matter most.
- +1.6Safety / Integrity: OSSF Scorecard 6.5 against 5.2
- +1.1Transparency: OSSF Scorecard 6.5 against 5.2
The SDK to extract data and interact with any site on the web. Get started with Claude Code, Codex, Eve, Mastra, and more.
Choose Stagehand if adoption matters most.
- +2.7Adoption: 1.6M weekly downloads against 296k
- A vs BEvidence coverage: 4 of 5 independent signal types, against 3
Where they differ
1 dimension identical: Identity 18.0 · Full evidence table
An independent, evidence-based trust comparison of agent-device and Stagehand, two Browser & Computer Use projects in the HVTracker registry. Scores come from public, checkable signals — supply-chain provenance, OSSF Scorecard, maintenance, and adoption — not popularity.
Full evidence
| Signal | agent-devicecallstack/agent-device | Stagehandbrowserbase/stagehand |
|---|---|---|
| HVTrust score | 91.9 | 89.2 |
| Evidence grade | A | A |
| Coverage grade | B | A |
| Overall rank | #12 | #35 |
| Rank in Browser & Computer Use | #1 | #5 |
| GitHub stars | 5.0k | 25.7k |
| Last updated | today | today |
| Build provenance | Yes | Yes |
| OSSF Scorecard | 6.5 / 10 | 5.2 / 10 |
| License | MIT | MIT |
| Downloads | 296k/wk | 1.6M/wk |
| Trust dimensions (points earned) | ||
| Safety / integrity / 25 | 20.6 | 19.0 |
| Identity & provenance / 18 | 18.0 | 18.0 |
| Transparency / 17 | 14.0 | 12.9 |
| Maintenance / 20 | 20.0 | 19.9 |
| Adoption / 20 | 16.2 | 18.9 |
| Runtime capability surface (full matrix) | ||
| MCP server | Implemented | Implemented |
| External providers | — | 3 — Anthropic, Browserbase, OpenAI |
| Requires API keys | No | Yes |
| Plugin surface | plugins | extensions |
| Provenance drift | Match | Match |
How to read this: HVTrust (0–100) weighs supply-chain signals (provenance, OSSF Scorecard, signed commits, open license) alongside real-world adoption, scaled by an evidence-confidence factor. Grade bands: A ≥ 80, B ≥ 65, C ≥ 50, D < 50. Signals refresh daily. Full methodology v4.4 →