# Playground

The **Playground** is an in-dashboard agent harness for testing MCP servers before wiring up a real client. Open it from **Connect > Playground** in the dashboard.

## Access requirements

<Callout type="info">
  Viewing this page requires the `mcp:connect`, `mcp:read`, or `mcp:write` scope. Running tools in the playground requires the `mcp:connect` scope, and editing saved playground or server configuration requires the `mcp:write` scope. The default [Admin role](/docs/ai-control-plane/org-admin/roles-and-permissions) includes full access; the default Member role can open and use the playground but cannot edit saved configuration.
</Callout>

## Testing a server

Pick any of the project's MCP servers — hosted and remote servers both appear in the picker. Authenticate against the server if it requires it, then chat with a model that calls the server's tools live.

While chatting:

- Mention tools directly to steer which ones the model uses
- Inspect the logs panel to see each tool call's request and response
- Adjust the configuration panel to change the server or session setup

## When to use it

Use the playground to verify a new source produces working tools, debug authentication before rolling a server out to real agents, and reproduce tool behavior reported by users without leaving the dashboard.