Orgo

@orgo list all workspaces and show me which computers are currently running in each one

Found 2 workspaces accessible with your Orgo key: **Production Workspace** (ws_abc123) - dev-server-01 (running, Ubuntu 22.04) - staging-db (stopped) **Testing Workspace** (ws_xyz789) - test-runner-03 (running, Python 3.11) - selenium-box (running, Chrome headless) This illustrative example shows how the MCP would enumerate your workspaces and their computers, including power state and basic configuration details.

This read-only operation requires a valid API key with workspace:read scope. Useful for discovery before targeting specific computers with actions. No side effects, but the response reflects your account state at call time—computers may start or stop between queries.

@orgo create a new Ubuntu computer in my Testing workspace with 2 CPUs and 4GB RAM, then start it automatically

Created computer **test-ubuntu-47** (cmp_new456) in Testing Workspace: - Template: Ubuntu 22.04 LTS - Resources: 2 vCPUs, 4GB RAM, 20GB disk - Status: starting (typically ready in 45-90 seconds) This illustrative example shows the MCP confirming the computer was created and is booting. The actual computer ID and startup time depend on Orgo's infrastructure load. Usage charges apply per the workspace's billing plan.

This write operation consumes workspace quota and may incur usage charges. The MCP calls create_computer only after you confirm the workspace ID and resource allocation. Always verify the target workspace before running—deletion is permanent and cannot be undone.

@orgo take a screenshot of dev-server-01, then click the 'Deploy' button in the center of the screen if you see it

**Screenshot captured** from dev-server-01 at 2025-01-15 14:32 UTC. The display shows a deployment dashboard with a green 'Deploy to Production' button centered at coordinates (640, 480). **Pointer action executed**: Left-click at (640, 480). The button changed state to 'Deploying...' and a progress spinner appeared. This illustrative example shows how the MCP can inspect a running computer's display and perform GUI actions—actual coordinates and UI state vary by application.

Pointer actions (click, drag, scroll) are applied immediately and may trigger irreversible side effects like form submissions or deployments. The MCP captures a screenshot first to confirm visual state, but you should verify the target computer and action before calling. Rate limits apply to screenshot and pointer operations.

When Orgo beats headless browsers for visual regression testing

A 3-person QA team needs to test checkout flows across five staging environments every sprint. Orgo wins here because you spin up a fresh Linux desktop, run Selenium or Playwright with a real GUI, and screenshot every step without fighting Docker display drivers. The Execute Code tool runs your test suite, Control Pointer handles edge cases where you need to manually click a modal, and Get Screenshot captures the failure state for the bug ticket. The trade-off: if your tests already run headless in CI and never need manual intervention, Orgo adds cost and latency you don't need. But if you're debugging flaky UI tests or need a throwaway environment for exploratory testing, the API-key auth and 11-tool surface make this faster than provisioning EC2 instances. Call Create Computer after the user confirms the workspace—it bills immediately.

Reproducing customer bugs in isolated sandboxes without local setup

A support engineer gets a ticket: 'the export button freezes on Firefox 128 with our enterprise SSO.' Orgo is the right call when you need to replicate the exact OS, browser version, and network config the customer reported, but your team doesn't maintain a lab of physical machines. You List Workspaces to find the shared support workspace, Create Computer from a curated Firefox template, Execute Code to install the SSO certificate, then Get Screenshot to confirm the freeze. The Delete Computer tool cleans up after the ticket closes, so you're not paying for idle VMs. The boundary: if your product is a CLI tool or API, Orgo's GUI-first design is overkill. If the bug involves screen rendering, third-party extensions, or OS-level quirks, this beats asking the customer for more logs. One API key covers the whole support rotation.

When you need a sandboxed GUI for agent tool-use research

A 2-person research team is fine-tuning an agent that books travel by navigating airline websites. Orgo is the move when your agent needs to learn click-drag-scroll patterns on real web UIs, not just API calls. You Create Computer with a browser template, your agent calls Get Screenshot to see the page, then Control Pointer to fill forms and Execute Code to verify booking confirmations in the DOM. The 11-tool surface gives you programmatic control over every input method without SSH or VNC overhead. The limit: if your agent only needs to call structured APIs, Orgo's per-minute compute cost is wasted. If you're training on visual tasks—form-filling, CAPTCHA-solving, multi-step workflows—the List Templates tool gives you reproducible starting states, and Delete Computer keeps your training runs cheap. This is cheaper than maintaining your own Selenium grid for agent evals.