Browser QA agent, as a service

gemini-cua runs Google’s Gemini Computer Use agent against real Chromium. Submit natural-language browser tests; get status, logs, and a recorded video back over HTTP — or invoke the same queue from agents via MCP and A2A.

Gemini Computer Use Playwright Chromium Job queue Video + logs

OpenAPI / Swagger

Interactive API explorer with request/response schemas for every endpoint.

MCP setup

Connect Cursor, Claude, or any MCP client to create and poll browser tests.

A2A setup

Agent Card discovery and JSON-RPC so other agents can delegate QA tasks.

What it can do

Navigate sites, click, type, handle cookies/consents, assert UI — from plain English
Queue many jobs; concurrency controlled by MAX_CONCURRENT_JOBS
Persist status, full agent logs, Gemini token usage / estimated cost, and MP4 videos
Optional API key auth via CUA_API_KEY

Quick start

curl -s https://gemini-cua.apps2.povio.dev/health | jq

curl -s -X POST https://gemini-cua.apps2.povio.dev/v1/jobs \
  -H 'content-type: application/json' \
  -d '{
    "query": "Open https://example.com and summarize the heading",
    "initial_url": "https://example.com",
    "highlight_mouse": true
  }' | jq

# poll until succeeded | failed | cancelled
curl -s https://gemini-cua.apps2.povio.dev/v1/jobs/<id> | jq

HTTP API (summary)

POST /v1/jobs — enqueue a browser test (202)
GET /v1/jobs — list jobs (limit, status)
GET /v1/jobs/{id} — status, summary, artifact links
GET /v1/jobs/{id}/log — plain-text agent log
GET /v1/jobs/{id}/video — MP4 when ready
DELETE /v1/jobs/{id} — cancel if still queued
GET /health — liveness / queue snapshot

Full schemas and try-it-out: /docs (Swagger) · /redoc · /openapi.json

Auth

If CUA_API_KEY is set, send Authorization: Bearer <key> or X-API-Key: <key> on API, MCP, and A2A RPC calls. /health and this docs hub stay open.