Read-only agent endpoint
Closed beta. Let an agent read your confirmed brand facts and your products, and nothing else.
The project owner turns the endpoint on from Technical SEO → Site files → For agents by creating a token. The token is shown once and only its hash is stored; it stops working after 180 days, creating another replaces it, and Turn off revokes it. The beta is off for the whole platform until it is opened.
POST /api/agent/<projectId>/mcp speaks JSON-RPC 2.0 with Authorization: Bearer <token>.
It answers initialize, tools/list and tools/call for two tools:
| Tool | Returns |
|---|---|
get_brand_facts | The confirmed brand facts, normalized. |
get_products | Products from the catalog, 50 per page with a nextCursor; descriptions cut at 500 characters; at most 100 KB per response. |
It never returns a fetched page, it cannot change anything, and there is no cart and no order. Limits: 30 calls a minute per token, 1,000 a day per organization and 1,000 a month per project (a limit answers 429). We keep the time, the tool, the size and a hash of each response, not the response.
Declaration files for agents (/.well-known/api-catalog, /.well-known/ai-catalog.json,
/.well-known/ard.json) are built from what you say your site offers, stored with your other
site files and checked the same way. Make them only for services
you really run; a site without an API or an MCP server needs none of them. The product does not claim
that being agent friendly changes how often AI answers cite you.