Start with an editable test project, an existing automation your team owns and a workflow administrator who can inspect its final records. This Studio guide connects the customer journey; use the chosen provider manual for its complete Studio or Code setup.

Prepare the request and result

Before adding a tool, agree with the workflow owner on:

  • The customer or employee action it may perform and the resources it may access.
  • Who supplies trusted identity and a stable request ID, and which details the customer confirms.
  • The exact response fields that mean completed, recorded, pending or failed.
  • The system where you will verify the result and reconcile a timeout or duplicate.

Use the existing provider manual for Make, Zapier, n8n, Power Automate, Workato, Boomi or MuleSoft. An account credential alone does not supply an approved business workflow.

Connect and configure the agent

1

Prepare the provider action

Follow that provider’s Studio instructions in the same test project. Obtain the deployed endpoint/scenario, its credential and schema, and a verified test request. Have its owner test both a permitted request and an unauthorized resource before exposing the action to customers.

2

Attach it where the request is handled

Open Agent → Global tools → + and select the approved packaged action, API tool or MCP tool. For an existing Playbook, add it to that Playbook’s tools instead. Keep credentials, destination accounts and trusted identity bound to their defaults; describe only the details the agent should collect. A standalone request test does not attach a tool.

Keep the trigger and trusted defaults on the attached action. Photo:
3

Explain invocation and result use

Write a Trigger for the actual customer request, including the confirmation you expect before a write. In the agent’s instructions, require the current result before announcing success. Distinguish a recorded callback from a completed phone call, or a pending approval from an approved request. Backend/provider logic must enforce access, validation and deduplication; these instructions guide the model.

4

Exercise the full conversation

Use Run with the provider owner’s safe test data. Backend-populated variables do not automatically reach Studio Run: in this isolated test project, use temporary fictional defaults or the test mechanism supplied by your developer, then clear them. Ask for the service, confirm collected details and inspect the emitted tool call, current result and agent answer. Verify the matching execution and record in the provider.

Check the failure path and launch

Repeat with a rejected resource, missing required field and provider failure. The agent must not expose another customer’s record, announce completion from a receipt, reuse an earlier result or automatically repeat an ambiguous write. For a Workflow that captures results, clear old result variables and branch on the current result; a missing response field can leave an earlier value unchanged.

Set the target Environment’s published credentials and publish the tested agent, then repeat the conversation from the actual published client. Verify the account and destination reached by that client, not only the editor’s Draft test. Use Handle pending automation requests when the response is a receipt rather than the business result.

Use the up and down arrow keys to select a result, Enter to open it, and Escape to close the search.