# Agent Middleware API — machine discovery This origin is the human design-partner site. It does not mint credentials or execute protected tools. Platform engineering, AI infrastructure, and security teams operating internal MCP tools are the buyers; autonomous clients consume the discovery contract below. The product is one thing with one name: a gateway between an agent and one paid MCP (Model Context Protocol) tool. "Boundary", "control point", and "trust plane" describe what the gateway does; they are not separate products. The SDK ships as `b2a_sdk` and its verifier as `b2a-verify-receipt`, where `b2a` is short for business-to-agent. ## Governed loop discover -> authenticate -> authorize -> invoke -> meter -> receipt -> audit -> govern ## Canonical API https://api.thisisatest.tech ## Bootstrap Fetch in this order: 1. GET https://api.thisisatest.tech/.well-known/agent.json 2. GET https://api.thisisatest.tech/llms.txt 3. GET https://api.thisisatest.tech/mcp/tools.json 4. GET https://api.thisisatest.tech/openapi.json Before assuming real side effects, fetch: GET https://api.thisisatest.tech/health/dependencies Inspect the JSON body: HTTP 200 alone does not mean every dependency is ready. Check `status`, the selected tool's dependencies, `simulation_modes`, and `enable_proof_surfaces`. Health never replaces authentication or permit checks. ## Authority Protected routes use the X-API-Key header. Live keys are operator-issued; there is no public self-serve key mint on this site. ## Published receipt proof The public proof page is: https://www.thisisatest.tech/proof/ It labels the evidence as self-issued live gateway proof, not customer traction. The exact portable receipt and matching key snapshot can be checked offline with the repository SDK verifier. The receipt and key files are public; the verifier ships in the source repository, which requires access (below). ## Try the complete loop locally No live credentials are required. The source repository is private during the design-partner phase: request access from the operator (security contact below) and the steps run as written. Two local paths, both self-contained: Drive it yourself (a real server, you hold the keys): 1. Clone https://github.com/PetrefiedThunder/agent-middleware-api (private — source access on request) 2. Run `make quickstart` from the repository root. 3. Follow docs/quickstart.md: mint your own wallet-scoped key, issue yourself a permit, invoke one governed tool, replay it, overspend it, and verify the signed receipt offline. Watch it prove itself (no server, one command): 1. Clone the same repository. 2. Run `make prove-trust-plane` from the repository root. Both use a throwaway local SQLite database and real FastAPI routes. They demonstrate permit -> invoke -> charge -> signed receipt -> replay without a second charge -> out-of-scope denial. These are local proofs, not production settlement or compliance claims. ## Humans and operators https://www.thisisatest.tech/ https://www.thisisatest.tech/proof/ (download a receipt and verify it offline) https://www.thisisatest.tech/compare/ (named alternatives, including the rows this loses) https://github.com/PetrefiedThunder/agent-middleware-api (private — access on request) ## Longer machine brief The product boundary, the vocabulary, and the claims deliberately NOT made: https://www.thisisatest.tech/llms-full.txt ## Security contact https://www.thisisatest.tech/.well-known/security.txt