ui-automation-control-plane

How-To: MCP Agent Review Loop

Use this page when you want to connect Codex, Claude Code, or another MCP-capable client to Prooflane’s deeper review layer without turning the product into a generic chat assistant.

What This Integration Is

• a read-mostly MCP workflow over the same runtime the UI already uses
• a way to inspect release briefs, similar failures, manual-gate context, and template feasibility from an MCP-capable client
• a truthful protocol story for Codex, Claude Code, or another MCP host

What This Integration Is Not

• not an official partnership claim
• not a hosted SaaS copilot
• not a write-capable approval channel for risky release decisions
• not a replacement for Stress Lab, Runs & Blocks, or Advanced Review

Best Fit

Use this loop after a result already exists.

Think of it like asking a well-prepared operator to hand you the clipboard, not like opening a second control room.

Prooflane still starts from:

1. Stress Lab for the target and experiment mode
2. Runs & Blocks for the latest result and manual-gate state
3. Advanced Review for governed comparison

MCP comes in after that result path exists.

The Fast Review Loop

1. uiq_run_overview
2. uiq_generate_release_brief
3. uiq_find_similar_failures
4. uiq_list_manual_gates
5. uiq_explain_template_feasibility

What this gives you in plain English:

• uiq_run_overview tells the client what run it is looking at
• uiq_generate_release_brief drafts the same governed summary the UI exposes
• uiq_find_similar_failures finds historical failure cases that resemble the current run
• uiq_list_manual_gates shows whether a paused run still needs operator help
• uiq_explain_template_feasibility explains whether a reusable journey is a good fit for another target family
• The Command Center now drafts a copy-ready agent handoff prompt in Advanced Review, so a human operator can paste the same governed run context into Codex, Claude Code, or another MCP-capable client without rewriting the brief manually.

Why This Is A Truthful Claim

These capabilities are backed by repo-owned surfaces today:

• ReviewBoard loads an AI release brief and similar failures
• Runs & Blocks exposes a Manual Gate inbox and report surface
• MCP registers:
  • uiq_generate_release_brief
  • uiq_find_similar_failures
  • uiq_list_manual_gates
  • uiq_explain_template_feasibility
• MCP resources expose:
  • uiq://review/latest-release-brief
  • uiq://manual-gates/inbox-summary

So the truthful statement is:

Prooflane works with Codex, Claude Code, and other MCP-capable clients through a governed, read-mostly MCP surface.

That is very different from claiming a native chat product or a hosted AI assistant.

Minimal MCP Config Reminder

{
  "mcpServers": {
    "uiq": {
      "command": "pnpm",
      "args": ["mcp:start"],
      "cwd": "/ABSOLUTE/PATH/TO/REPO"
    }
  }
}

If you want the optional review-oriented tool groups, add:

{
  "UIQ_MCP_TOOL_GROUPS": "advanced,analysis,proof"
}

Where To Go Next

• Setup details: docs/how-to/mcp-clients-setup.md
• Fast command path: docs/how-to/mcp-quickstart-1pager.md
• Full tool contract: docs/mcp.md
• Product front door: README.md

This site is open source. Improve this page.