Test an unsaved tool
Runs an unsaved API or MCP tool once, the same way a call would, and shows the request and result. It makes a real request to the tool’s endpoint. Nothing is stored.
Scopes: tool:write
Authorizations
An API key from Settings → API keys, sent as Authorization: Bearer kb_live_…. A workspace key starts kb_; a personal key starts kbp_ and acts as its owner.
Headers
A unique key (a UUID works) for this request. Retrying with the same key returns the first answer instead of running twice.
255The API version to answer as, a date (YYYY-MM-DD). Leave it out for the current version (2026-11-07). An unknown version is refused with 400 unsupported_api_version.
2026-11-07 "2026-11-07"
Body
An unsaved tool, run by the same executor a call uses.
The body of an kind='api' tool (see services.tool_execution for templating).
- ApiToolSchema
- McpToolSchema
api, mcp_server 1 - 801 - 64Response
Successful Response
What a test run did. Which fields appear depends on the tool's kind.
True when the tool worked.
An API tool's HTTP status, or an MCP server's connection status.
An error code when it failed.
A readable explanation of a failure.
The arguments the tool takes (JSON Schema).
The HTTP request sent, with secrets masked.
The raw response, with secrets masked.
What the tool returned.
The text the agent would read.
{duration_ms}.
The MCP tool that ran.
What the MCP server says about itself.
A fingerprint of the MCP server's tool list (for approve-tools).
The MCP server's tools, when listed.
A Google Sheets tool's tab.
A Google Sheets tool's header row now.
Columns whose header changed since the tool was saved.