Skip to main content
Version: v4

Build an AI agent as a Verifiable Service

An AI agent is a service with a model inside. Make it verifiable the same way: a VS Agent next to it holds the DID and the credentials, and the AI agent exposes its own protocol endpoint behind trust resolution.

Steps​

  1. Deploy a VS Agent for the AI agent and make it trusted, as a standalone organisation or as a child of your organisation agent (child services). Its ECS Service credential describes the AI agent: name, type (for example AI_AGENT), description, operator.
  2. Declare the endpoint. Add an MCP or A2A service entry pointing at your agent's endpoint (endpoints).
  3. Gate the endpoint. Your endpoint accepts a token your backend issues over DIDComm to peers whose DID the VS Agent resolved as trusted, or it accepts a presentation of a credential your ecosystem defines for agents. Either way, trust is established before the first tool call.
  4. Verify peers before you call them. When your AI agent chooses a counterpart, resolve its DID through the indexer (or through the Verana MCP server's tools) and proceed only on trusted; find candidates in the trust graph by what they prove.
  5. Identify the humans. Issue ECS Badges to the people who operate or supervise the agent, from your organisation agent (the Badge schema is open to issuers and permissionless for holders). A counterpart that needs a human in the loop asks for one.

Example​

The mcpi-two-vs-agents repository runs two AI agents, each with a VS Agent, that verify each other before exchanging anything.

Next​