Skip to content
AssureAgent

Connect AssureAgent to your AI

AssureAgent exposes Vibe — your conversational QA developer — as a Model Context Protocol server. That lets Claude Code (today), Claude Desktop / Claude.ai (next), and ChatGPT (coming) talk to Vibe like a colleague: design scenarios, run tests, grade outcomes — all from where you already work.

Why MCP, and why one big tool?

Two shapes are common when exposing a platform to LLMs:

  • A big schema of granular tools — create_scenario, run_test, get_report. Easy to introspect; verbose to use.
  • One conversational tool — talk_to_vibe(message). Elegant, but the external LLM can’t introspect capabilities by reading a schema.

We picked the second shape plus a few deterministic shortcut reads. Vibe already understands the platform end-to-end (it drove Manager before you got here). Handing it a message and getting back a smart reply is closer to how a real QA engineer would work.

What tools ship today

talk_to_vibe Primary agent-to-agent tool. Vibe answers in prose; hand it any request.
poll_conversation Async companion — long-running work like test runs pause and come back.
list_projects Deterministic read: your workspaces.
list_recent_runs Deterministic read: recent voice + chat runs with verdicts.
list_scenarios Deterministic read: scenarios in a project (or all).

Setup

All three clients are supported today via a personal aa_live_ API key. Public marketplace listings (OAuth 2.1) are next.

Claude Code One-line `claude mcp add` on the CLI.
Claude Desktop & Claude.ai Edit config.json (Desktop) or add a custom connector (web).
ChatGPT Team / Enterprise custom connector with Bearer auth.

How it counts

Every MCP conversation uses your AssureAgent turn allowance — same bucket as the portal chat. Your AI conversations tab (Manager → AI conversations) shows every conversation that came in through MCP, badged by which client drove it (Claude Code / Claude / ChatGPT), so you can see where usage went.

Verify your setup

The MCP server exposes a small public metadata endpoint so you can confirm the endpoint is reachable and see what tools are live without touching your API key:

Terminal window
curl https://mcp.assureagent.ai/mcp/info

Returns service metadata + the current tool inventory (name and description). No key required. If this comes back cleanly, your MCP client (Claude Code / Claude Desktop / ChatGPT) just needs your aa_live_… key to start driving Vibe.

Rate limits

Each API key gets 120 requests per minute on /mcp (per Cloud Run instance). Responses include X-RateLimit-Limit and X-RateLimit-Remaining headers. If you hit the cap, we return 429 Too Many Requests with a Retry-After header — your LLM client should back off and retry. Long-running Vibe work (test runs, path generation) doesn’t burn against this limit; the poll cadence does.

Security

  • Auth is a personal AssureAgent API key (aa_live_…) — the same identity you have in the portal. Each engineer configures their own.
  • Vibe never operates with elevated privileges. If you can’t do something in the portal, Vibe can’t do it for you either.
  • MCP calls are TLS-terminated at mcp.assureagent.ai and validated against the auth service before they reach chat / Vibe.