GateCore

GateCore is the governed transaction layer for AI agents. Every agent request is identity-verified, policy-checked, priced, settled, and receipted on a tamper-evident ledger before it touches tools, data, or money.

This repository is GateCore's public integration surface: the MCP server manifest, a quickstart, runnable client examples, and the agent-facing capability file. It is the canonical reference for the com.gatecoreai/marketplace entry in the official MCP registry.

What you can do today

Point any MCP client at https://mcp.gatecoreai.com/mcp and, with no key and no account:

• Discover governed capabilities. discover_listings returns published marketplace listings with machine-readable price, required scopes, and minimum trust terms attached.
• Read the contract before you commit. get_listing returns one listing and its machine-enforceable contract: price, required scopes, minimum trust, target, and data classification. Your agent can evaluate the terms programmatically instead of parsing marketing copy.

With a GateCore MCP access key (gcmk_ prefix), an agent can also:

• Request a governed procurement decision. procure returns PROCURE, REVIEW, or DENY. An eligible decision carries gateway delegation metadata, and execution happens through GateCore's signed gateway API.
• Read its own audit trail. list_procurements returns the procurement decisions visible to the credential's tenant.
• Submit a consented lead. submit_lead delivers a qualified consumer lead to a listing that accepts them.

QUICKSTART.md has the exact handshake and copy-pasteable calls. examples/ has short, dependency-light clients for discovery and listing fetch.

Why terms are machine-readable

An agent that wants to buy something on the open web has to read a human sales page, guess at the price, guess at the scope of what it is allowed to do with the result, and take the seller's word for quality. GateCore replaces all three guesses with structured fields on the listing:

• Price is stated in cents on the contract, so an agent can filter by budget before it ever asks to transact.
• Scope is the explicit set of permissions the capability requires, so an agent can check the request against its own policy instead of discovering the mismatch after the fact.
• Trust is a minimum threshold expressed on the contract. Trust on GateCore is earned from verified outcomes. It cannot be asserted by the seller or purchased.

Because those terms are attached to the listing rather than to a sales page, the decision to transact is one an agent can make on its own and a human can audit afterward.

Receipts and the audit trail

Every governed decision produces a signed receipt that binds the request, the policy decision, and the settlement together. Receipts are Ed25519-signed and independently verifiable: GateCore publishes the receipt signing keys at https://api.gatecoreai.com/v1/receipts/keys, and a client-side verifier at gatecoreai.com/verify recomputes the canonical bytes and checks the signature entirely in the browser. Verification requires no account and no call back to GateCore, so a receipt stands as evidence to a counterparty who does not trust us.

That is the point of the design. The decision is bound to the evidence, so "what did the agent do, under whose authority, at what price, and on what terms" has one answer that survives the session.

Public endpoints

Registry entry

GateCore is listed on the official Model Context Protocol registry as com.gatecoreai/marketplace. server.json in this repository is that manifest.

Access keys

Discovery is open. Procurement, procurement history, and lead submission require a GateCore-issued MCP access key so that agent identity and tenant scope come from the credential rather than from tool-call arguments. Get one self-serve, free, at app.gatecoreai.com/developer/portal. For anything else, contact hello@gatecoreai.com.

License

Apache License 2.0. See LICENSE and NOTICE.