Discovery pack · Agent builders and API maintainers
Ship an AI agent integration
A seven-check launch path for an API or MCP server: verify health, OpenAPI, MCP tool discovery and x402 payment metadata before giving an autonomous agent authority.
AI agent integration checklistMCP readinessOpenAPI auditx402 server auditAPI health
Direct answer
What should be verified before an AI agent uses an API or MCP server?
To ship an AI agent integration, verify four layers in order: transport health, the machine-readable API contract, MCP capability and authorization, then x402 payment and discovery metadata. Run the seven bounded checks below before client testing. A passing result is readiness evidence, not production certification.
Buyer proof
A payment challenge is not a delivered result.
Payment authorizes the intended request; it does not prove the declared result arrived. Before retrying or handing output to another agent, retain the provider selected, the unit and maximum charge, the accepted request, and the response evidence.
DeliveryMatch the response to the declared schema, freshness, request or correlation ID, and the artifact or receipt hash.
FailureStop retries when status or idempotency is unknown; never infer delivery from a 402 challenge, settlement receipt, or HTTP 200 alone.
RecoveryKeep the provider refund or support path and the status query needed to reconcile a timeout or partial result before automation continues.
Boundary: This is bounded delivery evidence, not a security, reliability, ranking, or production-readiness certification.
Readiness workflow
The four-layer launch path.
Run these layers in order. Each one narrows a different class of integration failure; none replaces client-level acceptance testing.
01
Verify transport health
Confirm that the public endpoint is reachable, returns predictable HTTP status and content types, and fails within a bounded timeout.
util_api_health_report · util_x402_server_probe
02
Inspect the machine-readable contract
Compare the published OpenAPI description with the paths, request shapes, responses and authentication behavior an agent will actually use.
It produces bounded evidence about endpoint health, OpenAPI shape, MCP readiness and x402 resource metadata at the time each route runs. It does not certify the whole system.
Do I need both OpenAPI and MCP?
No. Use the checks that match the surface you ship. OpenAPI describes HTTP APIs; MCP exposes resources, tools and prompts to compatible clients. A service can publish either or both.
Is a passing readiness score a production certification?
No. Production acceptance still requires client-level tests, authorization checks, failure handling, observability and a review of the actions the agent is allowed to take.
Will this pack list my service in a directory?
No. It can expose issues that block discovery, but indexing and ranking belong to each directory or registry and may happen asynchronously.
How is the pack priced?
There is no bundle fee. Each check is an independent pay-per-result route with its current USDC price shown below, and every AgentCash command includes a maximum amount.