A local MCP server that gives Claude Code, Cursor, and Codex real hands on your iOS Simulator and Android emulator — drive the UI, map the whole app, snapshot state, and watch backend errors live. 66 tools. Free and open source.
First 25 only · early access now + 50% off hosted for life · refund anytime
Recorded by proberun's own start_recording tool · dual-screen edit by proberun-tracecast · both ship in the repo
You describe the test in plain language. Claude reads an indexed accessibility tree — not pixels — and calls the right tools in sequence. It waits between screens, logs its reasoning, and stops the moment something regresses.
What you type
Why it's reliable
What runs on your Mac
Most mobile test tools re-discover the UI every run and treat the backend as a black box. Proberun is built for speed, debuggability, and AI agents.
Proberun autonomously walks your whole app on every build — mapping each screen and transition into a graph. Tests navigate by name instead of re-discovering the UI, cutting tokens ~80%.
save_state / restore_state via simctl clone. Restore a logged-in state in 2–5s instead of a 30–60s cold boot + re-login. Playwright contexts, for iOS.
Capture the app's logs and full HTTPS traffic during a test. Auto-classify Firebase, Supabase, Stripe, and Sentry errors. See the whole causality: tap → request → 401 → broken screen.
When the accessibility tree is sparse, Apple Vision OCR (free, local) or a vision LLM reads the screen — so RN, Flutter, Unity, and canvas UIs still work.
Every run records reasoning, tool calls, screenshots, and a video. proberun-tracecast renders a dual-screen replay you can scrub — the demo on this page was generated by it.
No HTTP adapter, no CLI wrapper. Claude Code, Cursor, and Codex call the tools directly with full context. The AI is the test runner.
Proberun captures the app's logs and network traffic during a test and classifies the errors. When a flow breaks, Claude knows whether it was a Firebase 401, a Stripe decline, or a UI bug — without you digging through Xcode.
Identical tool shapes across iOS today and Android now. Your AI editor calls them directly.
One server. iOS Simulator + Android emulator, same tool names. Read the docs →
Local is free forever. When you need parallel runs in CI, point Proberun at our cloud — your tests execute on hosted simulators and real devices, results stream back with full traces. You only pay for the minutes you run.
We won't pretend to beat mature tools on real-device fleets today. We're open about what ships now and what's coming.
| Capability | Proberun | mobai.run | Maestro | Appium |
|---|---|---|---|---|
| Free local OSS | ✓ | — | ✓ | ✓ |
| MCP-native (Claude/Cursor) | ✓ | ✓ | — | — |
| App Atlas auto-map | ✓ | — | — | — |
| State snapshot / restore | ✓ | — | — | — |
| Backend log + network capture | ✓ | — | — | — |
| Vision OCR fallback | ✓ | ✓ | — | — |
| Trace replay video | ✓ | — | — | — |
| iOS support | ✓ | ✓ | ✓ | ✓ |
| Android support | ✓ | ✓ | ✓ | ✓ |
| Real devices | soon | ✓ | ✓ | ✓ |
| Entry price | $0 | $5–10/mo | $0 / $250 cloud | $0 + infra |
Spot something unfair? Open an issue — we'll fix the table.
What's done, what's next, and what we're exploring. The changelog lives in the repo; the direction lives here.
The local tier is never gated. Every moat — Atlas, snapshots, observability, vision — runs free on your machine. Paid is for parallel cloud runs, history, and the team layer.
$49 once. Early access to the local tool now, 50% off hosted for life, a founder badge, and a direct line to shape the roadmap. Refund anytime before hosted ships.
Business ($499/mo · real-device farm) and Enterprise — hello@proberun.com
5 minutes to your first AI test
No. Built in the open by @DaltonTheDeveloper. Apache 2.0 from day one; the roadmap, decisions, and rough edges are all public. The local tier is genuinely free forever.
Sim + emulator today. Real-device support (signed WebDriverAgent) is on the roadmap for the Business tier. We won't pretend otherwise.
Native SwiftUI/UIKit and Android views work great. Sparse trees fall back to Apple Vision OCR (free, local) or a vision LLM — so canvas-rendered apps still work.
Anything that speaks MCP: Claude Code, Cursor, Codex. Tested primarily on Claude Code.
No. The local CLI runs entirely on your machine. Telemetry is opt-in and anonymous — tool names and error counts only, never code or arguments.
Same tool surface, but open source and free local, plus App Atlas auto-mapping, state snapshots, and full backend observability — none of which they have.
Free forever locally. Five minutes to install. Reserve the hosted tier if you want CI, parallel runs, and replays.