AI SDK MCP client close() in onEnd: Sume jobs keep running

Closing the MCP client in onEnd only ends your connection to Sume. A video job submitted through a tool call is a Sume job with its own id, so keep the job_id and read it later with jobs_status, or cancel it with jobs_cancel.

Client facts are from the AI SDK MCP page; job behavior from Sume's Jobs and results and MCP tools and gates, read 2026-09-30.

When does the AI SDK say to close the client?

The page says to close based on your usage pattern: for short-lived use, close in the onEnd callback of a streaming call; for long-running use, close when the application terminates.

import { createMCPClient } from "@ai-sdk/mcp";

const mcpClient = await createMCPClient({
  transport: {
    type: "http",
    url: "https://mcp.sume.com/mcp",
    headers: { Authorization: `Bearer ${process.env.SUME_API_KEY}` },
  },
});
const tools = await mcpClient.tools();
// ... streamText({ tools, onEnd: async () => { await mcpClient.close(); } })

What survives after close()?

What should I store before closing?

Store the job ids returned by the create tool. The docs' guidance is to re-read with the same ids, never to resubmit the create.

Should I use a long-lived client instead?

For a server that handles many requests, the page's long-running pattern applies: open the client once and close it when the app ends. The same rule holds either way: job state lives on Sume, not in the client.

What do I do with a job I no longer want?

Cancel it. jobs_cancel is a write tool and needs an idempotency_key, so the session must have mcp:write or an API key. Cancellation succeeds only before generation work starts; after that the job runs to completion. If the render is already done, jobs_result returns the artifacts, and the docs recommend reporting Sume public ids and media.sume.com URLs rather than pasting signed URLs, tokens or keys into chat logs.