The playground

Who this is for: Owner — requires integration.mcp.connect. Where: Integrations → Playground

The playground runs real calls against your own hotel, from inside the product, without you creating a key or writing a line of code. Use it to see what an endpoint actually returns before you build against it, and to check that a permission set does what you think it does.

Everything it does is live. There is no sandbox behind it — a read reads your data.

Issue a scratch key#

1. Select the permissions the key should carry.
2. Select Issue scratch key.

A key with a countdown on it. Fifteen minutes, the permissions you ticked, and integration.mcp.connect added for you.

The key is an ordinary API key with two differences: it expires fifteen minutes after it is minted, and the page keeps it for you so the buttons below can use it. integration.mcp.connect is added automatically, because without it the MCP calls on this page could not run.

Like any key, it is intersected with your own permissions — you cannot mint yourself something you do not already hold.

Mẹo

Tick only what you are testing. The point of the scratch key is not convenience, it is that you can reproduce a narrow key's experience without minting a real one. Ticking everything tells you nothing you did not already know.

Select Issue another to replace it with a differently-scoped key. The old one stops working when it expires; there is nothing to clean up.

Run a call#

1. Find the endpoint under Public API.
2. Select Run.

The status code and the response body appear under the row.

The same key, two endpoints, two answers. The scratch key held frontdesk.reservation.read and not frontdesk.guest.read, and that is the entire difference.

The request shape comes from the API contract, so there is no URL to type and no chance of testing a path that does not exist. Each row shows the permission it needs, so a 403 is never a mystery — compare the chip on the row against the scopes on your scratch key.

That pairing is the most useful thing on this page. Before you decide what a key for a partner should carry, mint a scratch key with exactly that set and run down the list. The endpoints that answer 200 are what you are about to grant.

Call an MCP tool#

1. Find the tool under MCP tools.
2. Edit the JSON arguments in the box, if the tool takes any.
3. Select Call.

The response is the raw MCP result — the same body an agent would receive, including the confirmation prompt from a sensitive tool. See MCP agents for what that prompt means and what to do with it.

Ghi chú

The tool list on this page is the full catalog, not the filtered view. It shows every tool with the permission each needs; what a connected agent sees is filtered down to its key. To see the filtered list itself, call tools/list with the key — the playground shows you the catalog and then refuses the calls your scratch key cannot make.

Submit a job#

Where the job surface is switched on for your organisation, the playground also runs jobs — the intent-shaped unit an agent is actually handed.

1. Pick a job.
2. Fill in the form. It is generated from the job's own declared input schema, so there is no URL to hand-type and no way to send a shape the job would reject.
3. Submit. You get back a handle rather than a result.
4. Select Refresh handle to read where the run got to.

That last step is the one worth doing by hand at least once, because it is exactly what a reconnecting caller does: a job is durable and asynchronous, so a real integration submits, goes away, and comes back to the handle. Watching it in the playground is cheaper than discovering the shape of it in your own client.

A sensitive job asks for confirmation here just as it does over the API.

What's next#

• API keys — mint the real, long-lived key once the scratch key has told you what it should carry.
• MCP agents — connect an agent to the jobs and tools you just called by hand.