Which Sume MCP tools to leave undeferred when Claude searches tools
Anthropic says to keep 3 to 5 frequently used tools non-deferred. Which Sume hosted MCP tools to pin and which to defer.

When you load Sume's hosted MCP server behind Claude's tool search, leave tools_list, tools_schema, jobs_wait and jobs_status undeferred, and defer the rest. Anthropic's tool search page says to keep your 3 to 5 most frequently used tools non-deferred so Claude can call them without searching first, and that at least one tool must stay non-deferred or the request returns a 400. The four named here are the ones almost every Sume run touches.
Why those four
tools_list lists the tools the session can see, along with safety metadata, and tools_schema returns one tool's contract by name. Together they are Sume's own discovery pair, and an agent that starts every task with them rarely needs a search. jobs_wait and jobs_status are what the agent calls after any submit, and a poll step that first has to search for its tool wastes a round trip.
The creation tools vary by task, so defer them. A task that makes an image does not need the avatar tools loaded.
| Sume tool | Role | Pin or defer |
|---|---|---|
| tools_list | List visible tools and safety metadata | Pin |
| tools_schema | Fetch one contract by name | Pin |
| jobs_wait | Hold up to 55 seconds for a job | Pin |
| jobs_status | Read job state | Pin |
| generate_image, avatars_create | Paid creation tools | Defer |
| assets_create, crawl_site | Write tools | Defer |
The two vendor rules to respect
Anthropic's page shows that per-tool configs override default_config, which is the mechanism for the pin.
- A deferred tool cannot also carry
cache_control; the API returns 400. Put the cache breakpoint on a non-deferred tool such as one of the pinned four. - Never defer the tool search tool itself.
- You still send every tool definition in the
toolsarray on each request;defer_loadingcontrols only what enters the context window. - With the MCP connector, set
defer_loadingin the toolsetdefault_configand override the pinned tools inconfigswithdefer_loading: false.
What does not change
Pinning is a context decision, not an authorization one. A read-only OAuth session still cannot run a paid or write tool, whatever Claude has loaded; the call returns insufficient_scope. Paid calls still need an idempotency_key, and max_spend_usd is still enforced only if you send it. For the Claude Code side of the same idea, see alwaysLoad false and deferred Sume tools.
Check the choice against your tasks
The four pins above suit an agent that submits and waits. A read-only research agent might pin tools_list, tools_schema and a read tool instead and never load a creation tool. A reporting agent that only reads jobs might pin jobs_status and jobs_result.
Anthropic's advice is to monitor which tools Claude discovers and refine from there. Log the search queries your agent makes for a week, and move any tool it searches for every run into the pinned set.
Sources
Related posts
More in Agents
- Haiku 5.5 effort for an agent that polls Sume jobs
Claude Haiku 5.5 defaults to medium effort. What that means for a loop that calls Sume jobs_wait, and when to move to low or high.
- Use a low-effort Haiku 5.5 subagent for Sume dry_run previews
Anthropic recommends low effort for subagents, and warns it can skip checks. Where a Haiku 5.5 subagent fits in front of a paid Sume call.
- Scheduled run spend cap: a per-call number can only lower it
A Sume schedule's default generation cap is $1.00. A per-run number is clamped to the lower of the request and the schedule cap, and 0 is rejected.
- Sume MCP avatar tools: 5 read-only and the paid create set
Hosted Sume MCP lists five read tools for avatars and avatar videos and a paid group for creates and previews. What OAuth mcp:read sees and the dry_run flow.
Written by Sume