Skip to content

Connect actions in your app

Open in ChatGPT Open in Claude

Defining an action in Agent Studio (details, inputs, response mock, and evaluations) covers the Studio side. To let the agent run the action, your front-end developer implements it in your application code.

flowchart LR
  subgraph PM["PM / AI Lead"]
      D[Define action<br/>in Studio]
      S[Set inputs &<br/>response mock]
      P[Publish version]
  end
  subgraph DEV[Developer]
      C[Pull action key<br/>& schema]
      I[Implement<br/>handler logic]
      W[Wire into<br/>product code]
  end
  D --> S --> P
  P -- "Hand off key,<br/>schema, expected output" --> C
  C --> I --> W

Split the work

In Agent Studio, you define what the action does and confirm the agent routes prompts and extracts schema fields.
In your application, a developer connects the action to real logic: API calls, database updates, navigation, and more.

Hand off three things:

Item	What it is
Action Key	The action’s unique identifier (e.g. `create_task`, `invite_user`).
Schema	The input fields defined in the Studio (e.g. `email`, `organizationRoleType`, `name`).
Expected Output	What the agent should perform, plus any data the agent should use in its chat response, such as deep-linking to a page.

Developers use these to wire the action into your product’s code.

Pull actions in over MCP

If your developers use AI coding tools like Cursor, VS Code, or Windsurf, your actions and schema are available over MCP. Developers pull your Studio-defined actions straight into their coding environment, which speeds up implementation.

Next steps

PMs / AI leads: finalize and Publish the action version you want live.
Developers: connect the published action in code.