Skip to navigation
Observe

Playground

Test what a Tool Pack can reach before you assign it to a Group

The Playground lets you chat against a Tool Pack as a test user. Calls hit the same MCP endpoint employees use, so you see what they will see.

The playground runs only as a test employee. Add one on Provisioning under Employees → Test. Your role must be able to view employees.

Run a test

  1. Pick a Tool Pack
  2. Pick a test user
  3. Send a message
  4. If a Connector isn’t authenticated, click Connect, finish OAuth, and the agent retries

Each tool call shows as a card with its status (Queued, Executing, Success, Failed, or Blocked by policy), parameters, and response. Calls are real: they reach the third party and count toward your tool call volume.

What to test

GoalHow
Group accessAsk for something at the edge of the pack. A missing tool shows as the model not calling anything.
Tool descriptionsIf the agent picks the wrong tool, edit the tool description override and re-run. Edits apply on the next message.
Input overridesCheck that call arguments contain the forced value from a tool input override
Security rulesSend a prompt with an email or card number. Look for Blocked by policy or a missing field, then check Violations and alerts.

Open in an AI client

Open in hands the Tool Pack and test user to Cursor, VS Code, Claude Code, Windsurf, or a custom client. It doesn’t carry the transcript.

The config in the Connect to MCP server dialog contains your organization’s API key. Treat it as a credential.