استدعاء الأدوات من التعليمات البرمجية

Discovery told you what exists. Now you actually invoke a tool, sending arguments and reading back the result from your own code.

Use session.call_tool() with the tool name and a dictionary of arguments. The names must match the inputSchema you discovered.

Your arguments dictionary should satisfy the tool inputSchema. Sending the wrong shape leads the server to reject the call.

The result carries a content list of blocks. Most tools return text, so you read the text off the first block.

A tool that fails sets isError to true on the result instead of raising. Inspect it so your client can react gracefully.

Newer tools may also return structuredContent, a typed object alongside the text, when the server defines an output schema.

Arguments come from your program, a user, or a model. The client simply forwards them as a plain dict to the named tool.

Like every network call, call_tool is awaited. The await pauses until the server finishes the work and the result returns.

Each invocation is a request-response pair. The session matches the reply to your request by id, so results never get crossed.

Some tools take time. A robust client sets a timeout so a hung tool does not freeze your whole program.

Calling tools from code means your program decides the order, combines results, and turns raw output into something useful for the user.

How do you run a discovered tool from client code?

You invoked tools end to end: call by name with valid arguments, read the content blocks, and check isError before trusting the output.