agent-device
By Callstack. Category: Mobile testing with AI. Homepage
Callstack's documentation describes agent-device as an open-source command-line tool, MCP server and Node.js API that lets AI coding agents inspect, control and verify mobile applications. Agent sessions can be saved as .ad replay scripts, which the agent-device runner replays and which can be exported to a subset of Maestro YAML. Documented targets include iOS, Android, HarmonyOS and TV platforms, macOS and Linux desktops, and a managed browser. It runs locally, with cloud devices through third-party providers. Agent setup is documented for Cursor, Codex, Claude Code, Windsurf, Cline and Goose. It is MIT licensed.
Description written from the vendor's documentation.
Status
Status: Active
Facts
| Fact | Value | Evidence | Source | Verified on |
|---|---|---|---|---|
| Status | Active | Documented | github.com | |
| AI capabilities | Not documented | Not documented | None found | |
| Test creation | Documented | oss.callstack.com | ||
| Test output | Proprietary runner | Documented | oss.callstack.com | |
| Surfaces | Documented | oss.callstack.com | ||
| Hosting | Documented | github.com | ||
| Pricing model | Documented | github.com | ||
| Free tier | Yes | Documented | github.com | |
| CI integrations | Documented | github.com | ||
| Self-healing | None documented | Documented | oss.callstack.com | |
| Agent integration | Ships an official stdio MCP server started with 'agent-device mcp', plus agent skills installed with 'npx skills add callstack/agent-device'; setup documented for Cursor, Codex, Claude Code, Windsurf, Cline and Goose | Documented | oss.callstack.com | |
| Open source | Yes, licence MIT | Documented | github.com |
Published prices
No prices are published.
Alternatives
Tools in the same category (Mobile testing with AI), ordered by the number of shared creation and surface values, then by name. Each line lists documented facts that differ from agent-device.
- Apptest.ai: Mobile testing with AI; no documented difference in output, creation, hosting or free tier
- Digital.ai Testing: documents output Mixed; creation by natural language, prompt to test; hosting cloud; status acquired
- Kobiton: documents hosting cloud; no free tier
- Loadmill: documents creation by natural language; hosting cloud, hybrid
- Optics Framework: documents creation by natural language
- Repeato: Mobile testing with AI; no documented difference in output, creation, hosting or free tier
Questions
Is this page a review of agent-device?
No. This page is a fact sheet. Each fact shows its evidence level, its source and the date it was checked. We do not rate or rank tools.
Does agent-device have a free tier?
Yes. agent-device has a free tier. This is documented, checked on 2026-09-24.
What test output does agent-device produce?
Output: Proprietary runner. This is documented, checked on 2026-09-24.
Is agent-device still available?
Status: Active, checked on 2026-09-24.
Sources
- agent-device homepage Opened
- callstack/agent-device README (GitHub) Opened
- agent-device LICENSE (MIT) Opened
- agent-device docs: Introduction Opened
- agent-device docs: Replay and E2E Opened
- agent-device docs: Agent setup Opened
- agent-device docs: Device clouds Opened
- agent-device on npm registry (latest 0.21.13) Opened
Spotted an outdated or wrong fact about agent-device? Send the source and we apply it at the next monthly update.