Codex MCP required = true: fail fast when the Sume server is down

Set required = true on the Sume entry in config.toml when a Codex run is useless without Sume's tools, such as a scripted job that must generate or inspect media. Codex's MCP page says required makes startup fail if that enabled server cannot initialize, so a failed Sume connection stops the run instead of letting the agent carry on without it. For an interactive session, leave it off.

Codex details are from Codex: Model Context Protocol, read 2026-10-02. The Sume endpoint is https://mcp.sume.com/mcp per the MCP overview.

What do required, startup_timeout_sec and the grace setting mean?

The page lists startup_timeout_sec (default 10 seconds) as the time for the server to start, required as the fail-at-startup switch, and a top-level mcp_optional_startup_grace_ms that controls how long Codex waits for optional servers when it builds the initial tool catalog. That grace defaults to 1000 milliseconds. Set it to 0 to wait for each server's startup_timeout_sec instead. The page adds that required servers still use their startup timeouts.

When should the Sume server be required?

Required fits runs where a missing server would silently change the job. A Codex task told to "make the product clip and report the job id" that starts with no Sume tools has nothing to call, and the model may improvise an answer. A hard startup failure is easier to spot in logs and in CI.

It does not fit an everyday session where Sume is one tool among many. There, a transient failure would block everything. Use the default optional behaviour and let /mcp show the state.

Does a longer startup timeout help with slow Sume jobs?

No. startup_timeout_sec covers the server starting, not a tool call. Slow renders are a tool-call problem. Sume's jobs_wait holds one call for at most 55 seconds, and defaults to 50 when timeout_seconds is omitted, so Codex's tool_timeout_sec (default 60 on the page) has to stay above that. On wait_slice_expired, call jobs_wait again with the same ids rather than resubmitting the paid create. Codex MCP tool timeout covers that setting.

What does the first tool catalog look like if the grace runs out?

The Codex page describes the grace as a wait while the initial catalog is built, but it does not say what happens to an optional server that connects after the grace. Treat that as unspecified. If your workflow depends on Sume tools being present on turn one, either mark the entry required = true or set the grace to 0, then confirm with /mcp and a call to Sume's read-only mcp_health tool.

Sume's docs do not publish a connection time, so measure on your own network before choosing a number.