T7
Verifiable AI agents
Agent identity, MCP, agent-to-service trust.
In the tracks: AI Agent Builder.
This module is an outline: the objectives, sections and assessment are agreed; the material is being written by the Showcase and Certification Working Group.
Outline. The material of this module is being written by the Showcase and Certification Working Group. The objectives, outline and assessment below are the agreed scope for program version 2026.1.
4 class hours. Assessed by: exam and lab.
Learning objectives
After this module you can:
- Give an AI agent a DID and the Essential Credentials so that it is a Verifiable Service in its own right.
- Make an agent resolve trust before connecting to a service and present its identity when asked.
- Use the Verana MCP server from an agent and expose a service to agents.
- Explain delegation: what a person's wallet, a business wallet and an agent each hold, and how a service tells them apart.
Prerequisites
C1, C2, C4, and the reduced T2 reading.
Outline
1. Why agents need identity
The problem of C1 seen from software; agent-to-service and service-to-agent trust; what a logo cannot do for a program.
2. An agent as a Verifiable Service
DID, Service credential with an agent type, operator credential; what the Proof-of-Trust of an agent looks like.
3. An agent as a user agent
Resolve before connecting; verdicts as data; refusing untrusted services; presenting the User Agent credential.
4. MCP
The Verana MCP server; registry and graph tools; VS Agent tools; authorization of an agent on a service.
5. Delegation
Personal wallet, business wallet, agent: who holds what; delegated credentials; limits and revocation.
Assessment
Knowledge exam and a lab: an agent with its own DID connects to a playground service, is trusted, and completes one flow through MCP.
Reading list
docs.verana.io: Build, AI agents and MCP. Verifiable Trust specification v4: Verifiable User Agents. The Eclipse M4P webinar recording (verana-research).