MCP 2026-07-28 resultType: Sume tool results omit it and still work
The 2026-07-28 MCP spec adds a required resultType, and clients treat a missing one as complete. What Sume's tool results carry and what never appears.

Sume's hosted MCP tool results carry no resultType field, and under the 2026-07-28 specification that is safe: the changelog, read on 2026-10-04, says clients treat a missing resultType as complete. A Sume result is the standard shape, a content array with text entries plus an isError flag. There is no input_required result anywhere in the Sume server code, so a client never has to run the new multi-round-trip flow against it.
What does the new field do?
The changelog adds a required resultType with two values, complete and input_required. The second belongs to the multi round-trip request mechanism (SEP-2322), where a server asks the client for more input before it finishes. List results also gain ttlMs and cacheScope, and tools/list order becomes deterministic.
What does Sume return instead?
| 2026-07-28 item | In a Sume result? | What a client does |
|---|---|---|
resultType | no | treats the result as complete |
input_required | never returned | nothing to handle |
ttlMs / cacheScope on lists | no | applies its own caching policy |
isError with text content | yes | reads the error text as before |
How do long jobs work without input_required?
Sume does not pause a call to ask a question. A create tool returns a job, and you follow it with jobs_status, jobs_wait and jobs_result, as the Jobs and results page describes. A jobs_wait call holds for up to 55 seconds per slice and takes up to 20 ids, so a slice that ends with jobs unfinished is expected, not an error.
Missing information is handled before the call, not during it. Paid and write tools need an idempotency_key, and dry_run=true previews admission and cost without submitting, according to MCP tools and gates.
What can still go wrong?
Two things are worth testing in a 2026-07-28 client. First, make sure it does not reject a result for lacking the new field; the spec says it should not. Second, a very large result returns an explicit oversize error rather than a truncated one, because the tool output budget is 256 KiB serialized. That error does not mean the job failed, so read the job with jobs_result before you retry.
Because there is no ttlMs, do not cache the tool list across deployments on Sume's behalf. Re-run tools_list after you change credentials, since an OAuth mcp:read session and an API key see different tool sets.
Sources
Related posts
More in Developers
- MCP deprecations, removal from July 2027: audit Sume client
An MCP 2026-07-28 release candidate deprecates Roots, Sampling, Logging, Dynamic Client Registration and HTTP+SSE, removal no earlier than July 28, 2027.
- MCP OAuth without DCR: client ID metadata documents and issuer checks
MCP 2026-07-28 deprecates dynamic client registration for Client ID Metadata Documents and requires clients to validate iss. What it means for Sume.
- Python MCP SDK 2.3 subscriptions=False: a Sume wrapper that pulls
MCPServer(subscriptions=False) turns off push subscriptions in the Python SDK. A wrapper over Sume jobs can do without them by waiting in bounded slices.
- Python MCP SDK v1 now gets security fixes only: use v2 for new clients
The Python MCP SDK v1.x line gets security fixes only; v2 added MRTR and the 2026-07-28 spec. What that means when you connect a client to Sume's hosted MCP.
Written by Sume