Agents

Connecting an agent

zeypm speaks MCP, so an agent can drive the real thing — reading and updating work through the same rules a person goes through.

What this is

Not a chatbot bolted on the side. A connection gives an agent a set of tools scoped to one project: it can read the work, create and update tasks, comment, and use the modules that project has switched on.

Setting one up

  • Open the project's MCP tab and create a connection.
  • Give the resulting URL to your agent — Claude Code, or anything else that speaks MCP.
  • Authorise it once. The connection is scoped to that project, not your whole organisation.

What it can and cannot do

  • It works inside one project. A connection cannot reach another.
  • It respects the same module rules you do — a module that is off has no tools.
  • Every write is attributed to the agent and appears in the history like anyone else's.

Attribution is the point. An agent's work is on the record, so you can see what it changed and when — the same as any team member. Nothing happens invisibly.

Keeping an eye on it

The MCP tab shows what the connection has been doing and what it has used, so an agent that is being noisier or more expensive than you expected is visible rather than a surprise on a bill.

Why it matters

A team using agents seriously ends up wanting them in the same system as everyone else, rather than a parallel world that has to be reconciled by hand. That is what this is for.