Cloud/Connectors/Playground
Playground
The tab where a connector stops being configuration and becomes something you can poke: the MCP Playground, embedded in the connector, running its AI Capabilities live from the browser with real responses.
Playground is the tab where a connector stops being a list of settings and becomes a thing you can actually use. It is the MCP Playground, embedded right inside the connector: paste a Connector URL (or open it from the connector itself) and every AI Capability the connector exposes becomes a tool you can run from the browser, with the live response shown back to you. After Dashboard to watch it, Credentials to identify it, Connection Tokens to hand out keys and AI Capabilities to decide what is on, Playground is where you prove the capabilities actually work before your agents ever call them.
Playground
Test this connector live in the MCP Playground.
list_issues
List issues in a GitHub repository, filtered by state.
Parameters
Configure tool inputs
On the free plan the Playground shows an Upgrade plan card; with a credentialed connector it connects on its own. Configure credentials first or it stays on Setup required.
The mockup is the live, connected workspace. Search the tools, open list_issues, fill the owner and repo and hit Run tool: the JSON response lands in the result pane and the call is counted in Session and Execution history on the right.
How it opens
The Playground is not free for everyone, and it is not open before setup. Two gates stand in the way, and the console shows each one plainly:
- Paid plan required. On the free plan the tab is a card, not a workspace: a flask icon, the connector name, the line "Test this Connector's AI capabilities live, right from your browser.", an Upgrade plan button with a lock, and the note that the Playground runs connectors live with a valid connection token, available on paid plans.
- Credentials required. If the connector has a credential schema that is not configured yet, the tab stays blocked with a lock, the title "Setup required" and a button that takes you straight to the Credentials tab.
On a paid plan with credentials already configured, the connector owns the Playground and connects on its own the moment you open the tab: no URL to paste, no token to copy. The card you saw on the free plan simply becomes the live workspace below.
The workspace, once connected
A status bar sits on top of everything you do. A green Connected dot, the connector name, its version and the host it is running on, the live tool count and a Disconnect button. Disconnect tears the live session down; the connector owns the playground, so nothing outlives the view. Below it sit three panes.
Pick a tool
The left pane is the tool list, every capability the connector exposes. Type in the search box and it filters by name and description as you type, exactly like the console. Each row carries an index number, the tool name in monospace and a one line description, plus a small dot if you have already run that tool: green for success, red for error. Open a tool and the center pane fills with it.
Read its capabilities
The center pane opens on the tool header: the name, the annotation chips the connector published for it, and the description. read-only (emerald) means the tool only reads, destructive (red) means it writes or changes something, idempotent (blue) means repeating it is safe. Below, the Parameters form is built from the tool's own schema: a text box for strings, a number field for integers and decimals, pill buttons for enums, a switch for booleans, and a required asterisk on the fields the tool demands. Hover the help icon next to any field for its description.
See the live response
Hit Run tool and the console invokes the capability against the real connector. While it runs you see the spinner; when it returns, the result pane shows a success badge and the raw response in a JSON viewer you can copy to your clipboard. If a required field is empty, the form tells you "This field is required." and will not run until you fill it.
The session at a glance
The right rail keeps two things for the whole session: Session metrics and Execution history.
- Session is four numbers: Calls (how many times you ran a tool), Success rate, Errors and Avg latency. The success rate starts good and turns amber if errors creep in; errors stay muted until there is one to show.
- Execution history is every run you made this session: an icon for success or error, the tool name, a short summary of the arguments you passed, the duration in milliseconds and how long ago. Click any row to re-run it, or clear the whole list with the trash button.
Why this screen exists
Your agents will call these tools at 2 a.m. with no one watching. Playground is the place to find out, at noon, that create_pr_comment writes to the wrong repository or that list_issues needs a state you forgot. You test the capability in the browser, read the real response, and only then let your AI near it, and because the skills you silence on AI Capabilities never appear here, the playground and the perimeter move together. The remaining tabs cover the rest of the build: shipping changes, the record of what ran, and the business side of the connector.
What comes next
The connector still has tabs to cover: Deployments to ship changes, Audit Logs and Log Streaming for the record, and Settings and Monetize for the business. Each gets its own docs page, one at a time, in the console's own order.