Repository navigation
feat(mcpserver): progress restarts a tool call's timeout, up to spec.progress.maxDuration - #1487
Merged
Merged
Conversation
A tool call whose caller sent a progressToken now runs under an idle budget: every downstream progress notification restarts spec.timeout, and MaxProgressCallDuration (1h) bounds the call as a whole. Calls without a token keep spec.timeout as their whole budget.
Published Helm chart
|
| Chart | muster-crds |
| Version | 5.39.7-r9d8ab99dt20261013100821h0a42648 |
| OCI reference | oci://gsoci.azurecr.io/charts/giantswarm/muster-crds:5.39.7-r9d8ab99dt20261013100821h0a42648 |
| Digest | sha256:62fcd29dc1874f2ca953984fd4c428238dd2e165a6a38b067067579ab278420f |
| Registry | public — gsoci.azurecr.io |
| Git catalog | giantswarm-test-catalog (index) |
Pull this chart
helm pull oci://gsoci.azurecr.io/charts/giantswarm/muster-crds --version 5.39.7-r9d8ab99dt20261013100821h0a42648Posted by architect-orb · build 35340 · commit 0a42648 · updated in place on every push
Published Helm chart
|
| Chart | muster |
| Version | 5.39.7-r9d8ab99dt20261013100821h0a42648 |
| OCI reference | oci://gsoci.azurecr.io/charts/giantswarm/muster:5.39.7-r9d8ab99dt20261013100821h0a42648 |
| Digest | sha256:278d8d022e24d2a7e34eb5902ebc11b88fd8717102051073b98f49719dffd100 |
| Registry | public — gsoci.azurecr.io |
| Git catalog | giantswarm-test-catalog (index) |
Pull this chart
helm pull oci://gsoci.azurecr.io/charts/giantswarm/muster --version 5.39.7-r9d8ab99dt20261013100821h0a42648Posted by architect-orb · build 35341 · commit 0a42648 · updated in place on every push
QuentinBisson
marked this pull request as ready for review
October 13, 2026 09:52
…ration Progress extends a tool call only on a server that sets maxCallDuration, and only the tool call: every other operation keeps a plain spec.timeout deadline. Replaces the global one-hour cap.
QuentinBisson
changed the title
…default one hour The cap now lives in a progress block and defaults to 3600 seconds, so every server gets progress keepalive without configuration and each can tune or disable it (a value equal to timeout extends nothing).
QuentinBisson
changed the title
3 tasks done
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
On a tool call whose caller sent a
progressToken, each progress notification from the downstream server restarts the server'sspec.timeout, up to a new per-server bound,spec.progress.maxDuration(seconds, fromtimeoutto 3600, default 3600). Every other operation (initialize, list, ping, a call without a token) keepsspec.timeoutas a plain deadline.What is the effect of this change to users?
A long-running tool that reports progress (marge's sweep, giantswarm/marge#242) is no longer cut at
spec.timeoutwhile it is still working, with no configuration. A server that goes silent is still cut one timeout after its last notification. A server owner can lower the bound, or set it equal totimeoutto turn the extension off.How does it look like?
A call cut at the bound fails with
no answer within the server's progress.maxDuration of 1h0m0s: context deadline exceeded; the idle cut keepsno answer within the server's timeout of …. Aprogress.maxDurationbelowtimeoutor above 3600 is rejected by the CRD (CEL) and bycore_mcpserver_create/update/validate.Any background context you can provide?
giantswarm/roadmap#4394: the Bot PRs page's sweeps outlast marge's 120 s timeout on gazelle. MCP Tasks would be the right primitive, but the Tasks extension is not in mcp-go yet (mark3labs/mcp-go#987), so progress is the stopgap. The MCP spec lets an implementation reset the timeout on progress and says it SHOULD keep a maximum regardless; it is per server because servers that report progress need different bounds. The
progressblock is where further progress settings go, and what gets deprecated once Tasks land.What is needed from the reviewers?
Whether a one-hour default is right for every server.
Testing
Do the docs need to be updated?
CRD, configuration and MCP tools references list the new field.
Should this change be mentioned in the release notes?