Atinamos Verification
Core Principle You can pay to be tested. You cannot pay to be trusted.
Explore verification

Atinamos Evidence MCP

AI agents can query selected published Atinamos verification evidence directly through the Atinamos Evidence MCP.

Live MCP endpoint: https://verify.atinamos.co.uk/mcp

Official MCP Registry identity: uk.co.atinamos/evidence

Registry version: 1.0.0 — status: active

Atinamos provides the evidence. The buyer owns the policy and decision.

The MCP is a read-only machine interface. It lets an AI agent inspect evidence about paid machine services before applying its own procurement or risk rules.

Important: MCP v1 is not the full Atinamos database

MCP v1 does not query the full private Atinamos registry or operational database.

It currently reads only the selected, sanitised evidence Atinamos has deliberately published. This is an intentional production boundary while the public interface, evidence semantics and buyer-policy behaviour are being proven in live use.

If MCP returns known: false, an empty search result or a policy result caused by missing published evidence, that does not mean Atinamos has never observed the service. It means that service is not represented in the current published MCP evidence corpus.

The private Atinamos catalogue is broader than the public MCP corpus. Private catalogue records, marketplace observations and unpublished verification material are not automatically exposed through MCP v1.

This boundary is deliberate. It prevents the public interface from silently exposing internal research, operational records or evidence that has not crossed the publication boundary.

As the interface is proven, the intended direction is for MCP to expose an approved public subset of the wider Atinamos evidence registry. Until then, MCP results should be interpreted as published-evidence results, not as an exhaustive search of everything Atinamos may hold privately.

Available tools

  • lookup_evidence(endpoint) — look up published evidence for a service route;
  • service_history(endpoint) — inspect timestamped published observations for a service;
  • search_services(query, limit) — search services represented in the published evidence corpus;
  • evaluate_policy(endpoint, policy) — evaluate published evidence against rules supplied by the calling buyer.

The policy result is eligibility under the supplied policy, not an Atinamos universal trust or safety verdict.

Example use

A buyer agent can discover a machine service, ask Atinamos what published evidence is held, then apply rules such as:

require known published evidence
minimum 1 successful fulfilment
maximum 0 failed fulfilments
require paid evidence
maximum evidence age 30 days
maximum observed price 0.01 USDC

The same evidence can produce different outcomes for different buyers because each buyer owns its own policy.

Read-only boundary

The public MCP does not hold wallet credentials, create or sign payments, trigger paid verification runs, write to the verification database, mutate evidence or purchase a service on behalf of the caller.

MCP v1 currently reads the published Atinamos evidence corpus. It does not expose every private registry or operational database record.

Developer documentation

Full public integration documentation, tool references, examples, policy schema and official Registry metadata are published in the Atinamos public technical repository:

Atinamos Evidence MCP documentation on GitHub

The MCP itself is the machine endpoint at https://verify.atinamos.co.uk/mcp; this page is the human-readable explanation.