Skip to main content

Understanding agents

View as Markdown

In EventCatalog, agents represent AI-powered components in your architecture. They sit alongside services, messages, domains, and flows as first-class catalog resources.

An agent typically wraps a large language model (LLM) and is given access to tools (MCP servers, APIs, databases) so it can take autonomous or semi-autonomous actions.

Why document agents?

As AI agents take on real responsibilities in production systems they become part of your architecture whether or not they are documented.

Cataloging them gives your team:

  • Discoverability — engineers and stakeholders can find what agents exist, what they do, and who owns them.
  • Model governance — the model block captures which provider, model, and snapshot version the agent runs on so drift is visible in the catalog.
  • Tool transparency — the tools array lists every external capability (MCP server, API, database) the agent can reach.