OPENAI WEBMCP CHALLENGE · LIVE IMPLEMENTATION
RetailAgentOS WebMCP Agent Storefront
The browser agent asks. RetailAgentOS checks what the retailer can actually promise.
WebMCP exposes only the next safe browser action while RetailAgentOS validates inventory, pricing, fulfillment, quote requirements, shopper constraints, and merchant policy.
Choose a native browser-agent mission or the guided replay.
Run the WebMCP demo
Try with a browser agent
Checking for native WebMCP…
- Open this page in a WebMCP-enabled browser.
- Give the browser agent this mission.
- Watch Mission Control update as tools are invoked.
Copying the prompt does not run the agent — it hands the shopper mission to whatever browser agent you connect.
Watch guided mission
No WebMCP agent connected? Watch the complete flow run through the exact same RetailAgentOS handlers, without an external agent.
Shopper & storefront
Mission Control
Not yet runningThe same shared descriptors are used whether an agent or the guided mission invokes them.
Safe next actions, right now
Checkout is never exposed
- Mission received · run the guided mission to see the shared descriptors act.
RetailAgentOS decision
The decision authority, not the agent, exposes the next action.
Evaluate the shopper mission first.
Why the storefront becomes meaningfully better with WebMCP
Ordinary browser automation
- Guesses from buttons and product text
- May act on stale inventory
- May invent substitutes, prices, or delivery promises
- May proceed beyond merchant policy
RetailAgentOS with WebMCP
- Uses typed storefront capabilities
- Receives deterministic merchant decisions
- Gets merchant-valid repairs
- Exposes only the next safe action
- Never receives a checkout capability in this showcase
Developer evidence
Developer evidence
fresh-cornerSession: fresh-corner-1Registration source: no native registration observedRegistry parity: Browser observation unavailableRetailAgentOS-tracked tools: 0Browser-observed tools: unavailableEach tool descriptor carries an exact name, JSON Schema input, annotations, and lifecycle (registered / invoked / completed / failed / cancelled / unregistered). Native and guided invocations call the identical descriptor handler and RetailAgentOS gateway; only the telemetry source field differs (native vs. replay).
Full event log (0)
[]
One decision, multiple channels
Browser-agent actions demonstrated on this page.
Shipped for this controlled showcase
Commerce discovery and transaction projection foundation.
Foundation in place
Remote/client integration path, designed separately.
Designed, not shipped
Human and machine projections of the same canonical facts.
Partial — not every channel is live
All of these derive from the same canonical merchant/catalog/policy objects and the same deterministic RetailAgentOS decision — no channel recalculates commerce policy on its own.