Sampling - mcp-use

When a server tool needs an LLM during execution, it sends a sampling/createMessage request to your client. Provide onSampling to handle it.

import { MCPClient, type OnSamplingCallback } from "@mcp-use/client";

const onSampling: OnSamplingCallback = async (params) => {
  const text = await yourLLM.complete(params);
  return {
    role: "assistant",
    content: { type: "text", text },
    model: "your-model",
    stopReason: "endTurn",
  };
};

const client = new MCPClient(
  { mcpServers: { demo: { url: "http://localhost:3000/mcp" } } },
  { onSampling },
);

await client.connect("demo");

Per-server override

Set onSampling on an individual server config to override the global default.

React

With useMcp, pass onSampling in options. With McpClientProvider, handle server.pendingSamplingRequests and call approveSampling / rejectSampling.

Reference

Runnable example: sampling-client.ts

⌘I