Docs navigation

Use your server

Playground & chat

Before wiring a server into an agent, the Playground tab lets you call any of its tools directly from the browser and see exactly what comes back - whether it's the first run of a new tool or a check that yesterday's change didn't break anything.

A form per tool

The Playground reads a server's tools straight from its live MCP endpoint - name, description, and input schema - and renders a form for whichever one you pick. Run it and the result comes back raw, as your server returned it.

What bounds a call

A handful of limits keep one runaway tool call from affecting anything else: the initial connection has 10 seconds to complete, a call has 10 seconds of idle time before it's considered stuck (reset every time the tool reports progress), and 60 seconds total regardless of how much progress it reports. A response over roughly 1 MB is truncated rather than sent whole, and each project is capped at 30 tool calls a minute - enough for interactive debugging, not enough to double as a load test.

Scrubbed before it's shown

Playground and chat results both pass through the same PII scrubber before they reach the browser - so a raw record with a real name, email, or address in it doesn't end up sitting unredacted in a debugging session or a pasted screenshot. See data protection for what gets masked and how to confirm it's working.

Chat

The dashboard's Chat tab is the same idea at a higher level: a normal conversation that calls your server's tools as needed, so you can sanity-check how they read to an agent, not just what they return to a form.