opsis-mcp

Opsis MCP server — pre-deploy launch checks with Playwright proof, run from your coding agent.

Before you deploy, your coding agent (Claude Code, Cursor, Windsurf) connects to this MCP server, derives your critical user flows from the codebase (real routes/forms/selectors — not URL guessing), runs each flow in a real Chromium browser with full evidence capture, and submits an aggregate launch report to your Opsis dashboard.

Codebase (read by YOUR coding agent)
  └─ agent derives critical flows (routes/forms/selectors from code)
       └─ opsis_run_flow_check       ← per flow; Playwright runs LOCALLY
            ├─ video / trace / screenshot / console / network evidence
            ├─ pass = flow worked (no console/network/step failures)
            └─ on failure: guard-railed safe-fix prompt for the agent
                 └─ opsis_submit_launch_report → dashboard pipeline stage
                      └─ verdict: "Launch-ready: PASS / N ISSUES FOUND"

Install

You don't install it directly — point your agent at it with npx. Grab the exact snippet (with your project's keys pre-filled) from your Opsis project's Settings → Pre-deploy checks (MCP) card.

Claude Code:

claude mcp add opsis \
  -e OPSIS_APP_URL=https://<your-opsis-domain> \
  -e OPSIS_PROJECT_ID=prj_xxxxxxxx \
  -e OPSIS_MCP_KEY=opk_xxxxxxxx \
  -- npx -y opsis-mcp

Cursor / Windsurf (mcp.json):

{
  "mcpServers": {
    "opsis": {
      "command": "npx",
      "args": ["-y", "opsis-mcp"],
      "env": {
        "OPSIS_APP_URL": "https://<your-opsis-domain>",
        "OPSIS_PROJECT_ID": "prj_xxxxxxxx",
        "OPSIS_MCP_KEY": "opk_xxxxxxxx"
      }
    }
  }
}

Configuration

| Env var | What it is | |---|---| | OPSIS_APP_URL | Your Opsis deployment, e.g. https://app.opsis.id | | OPSIS_PROJECT_ID | The Opsis project id (prj_…) | | OPSIS_MCP_KEY | The project's MCP key (opk_…, from Settings) |

Prerequisite: Chromium on the developer machine — npx playwright install chromium. Evidence is stored locally under .opsis/predeploy/ in the repo your agent runs from.

Tools

| Tool | What it does | |---|---| | opsis_launch_check_guide | How to derive flows from the codebase (read first) | | opsis_run_flow_check | Run ONE flow locally with Playwright; evidence + fix prompt on failure | | opsis_submit_launch_report | Submit all results → dashboard stage + verdict | | opsis_launch_status | Latest launch-check status for the project |

Just ask your agent, e.g. "Run Opsis launch checks against my dev server on http://localhost:3000." It calls the guide first, runs each flow, then submits the report.

Security

• The MCP server never reads your codebase — the host agent does. Only flow results (title, route, status, one-line summary) leave your machine; video/trace/screenshot evidence stays local.
• The opk_ key authorizes only this project's launch-report API — never PostHog data or the dashboard.
• fill values must be synthetic (the guide enforces this); summaries are redacted before they leave the machine.

License

MIT — see LICENSE.