OpenAI WebMCP Challenge · synthetic demo

One prompt. A truthful review handoff.

The agent handles the repetitive preparation. The person keeps the decisions that matter—and nothing is submitted behind their back.

This standalone page exposes five challenge-only tools over fixed fictional data. Production /generator has exactly two different page-level WebMCP tools; remote MCP is complementary and separate.

Try with a WebMCP agent

Ask once. Watch the same workspace change.

Find remote AI product roles in Canada, choose the strongest match, prepare the application package, and stop for my review.

The manual fallback calls the same five local handlers. It does not prove WebMCP discovery by the browser.

Checking whether this browser exposes WebMCP…

Confirmed facts only Exact revision checks No credential handling No hidden submission

Shared human + agent state

Application workspace

Ready
Revision 12
Profile 18 confirmed facts
Target AI product · Canada
Rule Review every application
  1. 1
    Load role fixturesThree fixed fictional listings
    Waiting
  2. 2
    Select strongest fixtureIllustrative synthetic score
    Waiting
  3. 3
    Prepare manifestSynthetic document-status markers
    Waiting
  4. 4
    Stop for human reviewVisible package, no submit
    Waiting

Visible synthetic workspace detail

No fixtures loaded. The demo makes no network request.

    Selected
    None
    Package
    Not prepared
    Next owner
    Agent may prepare; human must review
    Waiting for a WebMCP tool call or the manual demonstration.