Prove your agents can work across boundaries — safely — before your organisation is ready to let you.

A neutral space, outside anybody's perimeter, to build and compare what is actually possible when agents have to work with systems and other agents that belong to somebody else.

Guided, runnable paths. Each clones from a repository and runs on your own machine against the Lab's hosted infrastructure.

Nothing is runnable yet. Everything above is planned or in development. The catalog says so on purpose — a recipe is only marked runnable once somebody who has never seen it can clone it and get a result.

Before your first recipe

All setup

3 things to put in place once. You do not run these — the Lab does.

  1. Get a Lab accountSign up with any email address you can receive post at — including a personal one.
  2. Connect your agent to the LabPoint an MCP client at the Lab catalog and sign in, so your agent can find its own way around.
  3. Open the sample chat client — optionalA hosted chat client, bring your own model key — talk to the Lab before you have built anything.

What your agent calls

All targets

5 sample endpoints the Lab hosts. Most admit only the Agent Gateway, so you reach them through a governed path rather than directly. That is the demonstration, not an obstacle.