Tools and MCP

Define a local tool with a typed signature:


tool priority_for(severity: str) -> int:
    if severity == "high":
        return 1
    return 3
 
type Ticket:
    priority: int
 
def main():
    ticket: Ticket = analyze(
        issue,
        "Classify this ticket",
        tools=[priority_for]
    )

Watch a tool call before it runs:


result: Ticket = analyze(issue, "Classify this", tools=[priority_for]) on tool_call(name, args):
    print(f"calling {name}({args})")

args is a mutable dictionary. Editing it changes the arguments sent to the tool. Returning a string skips the tool and sends that string back to the model.

MCP servers


use mcp github as gh
 
result: str = analyze(issue, "Find related issues", tools=gh.tools)

Configure how the server starts in kora.toml. Each server is its own sink, and every request has a deadline. See Language reference for the manifest shape.