OpenCode MCP timeout: 5000 ms tools fetch vs Sume jobs_wait

In OpenCode, the remote MCP timeout key is a number of milliseconds for fetching the server's tools, and it defaults to 5000. It is a connect-time setting. A Sume video job is bounded by something else: each jobs_wait call holds for at most 55 seconds, and you call it again with the same ids until the job is terminal.

OpenCode's side is from its MCP servers page; Sume's side is from Jobs and results and the MCP quickstart, read 2026-09-30.

What does the OpenCode timeout key cover?

The OpenCode page lists the keys for a remote server: type (required, "remote"), url (required), enabled, headers, oauth and timeout. It describes timeout as milliseconds for fetching tools, default 5000. It does not describe it as a limit on how long an individual tool call may run, so do not read it as one.

For Sume, the url is https://mcp.sume.com/mcp. Per the quickstart, once connected you confirm the server by calling tools_list or mcp_health.

Which Sume limit applies to a long render?

Every remote HTTP POST /mcp call holds at most 55 seconds, with 50 as the default when you omit the value. When a wait returns wait_slice_expired, the docs say to retry jobs_wait with the same ids and never resubmit the paid create.

When should I raise the OpenCode timeout?

Raise it only if OpenCode fails to load the Sume tools at startup. Raising it will not help a render that outlasts a slice, because that is the jobs_wait behavior above. The server also reports listChanged: false in its capabilities, so the tool list should not change underneath a session once fetched.

What should the agent do when a wait slice expires?

Call jobs_wait again with the same job ids. If you start several renders, one batch wait with job_ids is preferred over one wait per job. A related walkthrough is Using Sume from OpenCode.