Using the tool console
The tool console on Configuration > AI > Tool console runs a single MCP tool directly, through exactly the same path the /mcp endpoint and the internal agent use. Testing in the console is therefore testing the real path, not a simulation of it.
You use it to check that a tool does what you expect: does get_slots_for_objects return the times you think it does, does filter_customers find the right customer, and what exactly ends up in the audit trail.
Running a tool
- Pick the tool you want to run under Tool. Write tools carry a warning sign.
- Fill in the arguments under Arguments (JSON). The description and schema of the chosen tool help you along. Leave the field empty and it runs with an empty object.
- Click Run.
- Under Result the raw outcome appears: the text the tool returns, whether it errored, and how long it took.
- Below that sits the matching audit trail row, with among other things the sequence number, the correlation id, whether it was a write action and the fingerprint of the key used.
Write tools are real
Tools such as add_booking, subscribe_to_event, submit_form, escalate_to_human and call_colleague perform real actions. A booking is then a real booking, and call_colleague actually rings someone. send_verification_code sends a real SMS or email too.
That is why write tools can only be run while Teqa support is logged in as this licence, and the console asks for confirmation first. Read tools may be run by anyone with the right menu_confmod_ai.
Recognisable in the audit trail
A run from this page is recorded with source admin_console, not as mcp. A test booking from the console therefore never reads as a real agent conversation — handy when you later want to work out where a booking came from.
