llms.txt
+44 −18
https://www.escrova.io/llms.txt
− > prices, fees and terms. It cannot buy, bid or sell anything here.− Read. Every call in the sections below is a plain read with no authentication− and no wallet.− Four signed reads are answered as well, for an agent with its own wallet:− - GET https://www.escrova.io/api/agent/auth/whoami (does your signature work)− - GET https://www.escrova.io/api/agent/funds (what your wallet holds)− - POST https://www.escrova.io/api/agent/funds/check (does that cover a basket)− - POST https://www.escrova.io/api/agent/compare (landed cost of a shortlist)− None of them buys anything. How to sign them is at− https://www.escrova.io/auth.md.− Every other signed agent request is closed on this deployment: a request that− would quote an order, place one, bid, list, open a dispute or sign is refused− with `AGENT_RAIL_CLOSED` and `retryable: false`, so do not retry it. A person can− still buy on the storefront.− ## The same reads, as MCP tools− Streamable HTTP, tools only, and stateless. Send each request as a POST; GET− None of these tools takes a wallet. The tools, and the endpoint each one calls:− The table above is the whole set on this deployment.+ > prices, fees and terms. It can prepare orders and listing drafts for a person+ > to review. Only that person approves and pays or publishes on Escrova.+ Browse without authentication. With a live grant from a person, an agent can+ prepare an order or listing draft for them to review. The person signs in on+ the approval page, approves the order and pays from their own wallet, or+ publishes the listing. No agent tool pays or publishes. Every shopper can+ use this flow on mainnet, by the founder decision of 11 November 2026.+ Grants here allow only buy (prepare only), sell (draft only) and addresses+ (read only). Delegated spending and server wallets remain closed. A signed+ request needs a fresh challenge bound to its method, path, body and grant id.+ See https://www.escrova.io/auth.md. Operator discovery, write and tool controls+ can reduce the advertised tools. With writes off, preparation stops and reads stay.+ ## Signed reads for the agent's own wallet+ These HTTP calls remain available without a grant. Each uses its own signed+ challenge, as described in https://www.escrova.io/auth.md:+ - GET https://www.escrova.io/api/agent/auth/whoami+ - GET https://www.escrova.io/api/agent/funds+ - POST https://www.escrova.io/api/agent/funds/check+ - POST https://www.escrova.io/api/agent/compare+ ## MCP tools for browsing and preparation+ Streamable HTTP and stateless. Send each request as a POST; GET+ MCP Apps (2026-01-26) show English product and checkout cards with text fallback; host locale is not yet used, and payment checks need fresh signed reads and cannot confirm payment with the current status response.+ The first eight tools need no wallet. The rest authenticate the agent or+ prepare work for the person to approve. The tools, and the endpoint each one calls:+ auth_challenge POST /api/agent/auth/challenge+ whoami GET /api/agent/auth/whoami+ get_delegation GET /api/agent/delegation/grants+ list_addresses GET /api/agent/addresses+ prepare_checkout POST /api/agent/checkout-intents+ get_checkout_intent GET /api/agent/checkout-intents/<id>+ create_listing_draft POST /api/agent/listings/drafts+ update_listing_draft PATCH /api/agent/listings/drafts/<id>+ get_listing_draft GET /api/agent/listings/drafts/<id>+ prepare_listing_publish POST /api/agent/listings/drafts/<id>/prepare-publish+ The table above is the whole set when operator controls allow preparation.+ The person gives the limited grant at https://www.escrova.io/delegation.+ For MCP, call auth_challenge with walletAddress, targetTool and targetArguments+ including delegationGrantId, then sign its message and send the tool the same+ arguments with walletAddress, nonce and signature. whoami uses no grant.+ prepare_checkout accepts store items and a saved addressId, or the person's