Rigworks
Private beta — early access available

Real test environments
for AI coding agents.

Give Claude, Cursor, or any MCP-compatible agent real Windows, Linux, and Android environments. Build, push, run, and iterate — with actual logs and screenshots, not guesses.

Start free trial See how it works
claude-desktop — MCP tool call
> User: Fix the Windows build bug, then verify it actually works.

→ rigworks.build_for_platform("windows")       ✓ 842 KB exe built
→ rigworks.deploy_to_test_env(artifact_id)     ✓ deployed to env-3f2a
→ rigworks.run_test_scenario({...})            ✓ ran 34s, 3 screenshots
→ rigworks.capture_logs_screenshots_metrics()  ✓ no crashes, UI rendered

✓ Verified: Build succeeded. App launched cleanly.
4
Platforms supported
~60s
Environment provisioning
MCP-native
Works with any agent
100%
Real OS execution

Close the loop between AI and reality.

AI agents currently guess whether their code works. Rigworks lets them prove it.

Multi-platform

Windows native apps, Linux/Docker, Android emulators, web browsers — all in isolated, real environments.

One-click provisioning

Sign up, pick a plan, click Launch. Your dedicated rig is ready in about a minute.

MCP-native

Works with Claude Desktop, Cursor, Codex, or any agent that speaks MCP. Add one URL + token.

Self-improvement loop

Build → push → run → collect logs/screenshots → feed back to the agent → iterate until it passes.

Per-customer isolation

Each environment runs on its own VPS. No noisy neighbors, no data leakage.

Real hardware

Not a sandbox. Hyper-V virtual machines, Android emulators, actual OS processes.

How it works

Four steps from signup to a self-improving AI agent.

01

Sign up and pick your plan

Email + password. Pick Hobby, Pro, or Enterprise based on the platforms and hours you need.

02

Click Launch Environment

Rigworks provisions a dedicated rig with the platform runtimes pre-installed: Hyper-V, Docker, Android SDK, browser drivers.

03

Connect your AI agent

Copy the MCP URL and token into Claude Desktop, Cursor, or any MCP-compatible client. One-time setup.

04

Let the loop run

Your agent calls build, deploy, run, and analyze tools. Real execution, real logs, real screenshots. Iterate until done.

Stop hoping your AI's code works.

Start verifying it, on real machines, in under a minute.

Create your account

No credit card required for 14-day trial.