Tools

Select a tool to see its description, input and output schemas and annotations. Tools marked destructiveHint are flagged and ask for confirmation before running. With an environment marked as production active, a tool call asks first, since a tool may change data; tools the server marks readOnlyHint run without asking.

Arguments: use the form generated from the input schema, or switch to raw JSON. Generate args asks the assistant for example arguments.
Execute: shows the content blocks (text, JSON, images), structuredContent, isError and latency.
Display: show the output Pretty (JSON as a tree, images inline), Raw (searchable text) or as rendered Markdown.
Save the result: the down arrow at the right end of the result's bar saves it as JSON, Markdown (when the result is text) or text, and Copy sits before it (Save and export results).
Results stay while the app runs: each tool keeps its last result, arguments and assertions, also when you switch tabs, servers or views.
Assertions: status: success|error, JSONPath checks against structuredContent, or the parsed text content.
Save as test…: writes the call with its assertions as a YAML test to tests/mcp/ (the toast has Run now and Open test file; History's right-click menu has the same for an earlier call):

yaml

name: Search Customer MCP Tool
type: mcp
server: customer-mcp
tool: search_customer
arguments: { customer_id: "123" }
assertions:
  - { type: status, expected: success }
  - { type: equals, path: $.customer.id, expected: "123" }

status: error matches only a tool-level error (isError: true). A connection or protocol failure counts as a test error, never as the expected error.

Tools ​

Tools