MCP Core Maintainer Meeting - August 26, 2026 #3316

Attendees

• Den Delimarsky (@localden)
• Paul Carleton (@pcarleton)
• Peter Alexander (@pja-ant)
• Caitie McCaffrey (@CaitieM20)
• Kurtis Van Gent (@kurtisvg)

Purpose: Work through caching-header compliance issues, restructure the security interest group and several working groups, and prep the Skills-over-MCP proposal for a vote.

Key Decisions

1. Meeting moves to the Model Context Protocol calendar

• @localden announced the recurring meeting will move to the MCP calendar to remove admission friction and automate note distribution, which previously reached only Peter, Den, Paul, and David.

Outcome: @localden to transition the recurring meeting to the MCP calendar.

2. Subscription listening issues in the new spec version: raised, no decision

• @kurtisvg reported multiple clients continuously sending empty subscription requests without subscribing, and Antigravity integrations failing completely when they hit unsupported methods.

Outcome: No decision recorded.

3. Cache headers, mandatory vs. optional: no decision, directory checks planned

• @CaitieM20 reported startup-time performance hits on new servers from server-discovery timeouts. The C# SDK team shortened timeouts and added caching as a workaround.
• @pja-ant noted several servers advertise the 2026 version without returning cache-scope and cache-TTL values, causing SDKs to fail entire requests, and proposed making those fields optional.
• @pja-ant and @CaitieM20 felt failing entire requests is too strict in practice given how often servers omit the headers. @kurtisvg emphasized keeping the compliance burden on servers, anchored in the spec and conformance tests, and cautioned that client-side workarounds for missing requirements could weaken the conformance tests as a signal.
• @pcarleton and @localden suggested pragmatic client fallbacks: treat a zero TTL by using last retrieved data, and relist on errors.
• @pja-ant acknowledged full compliance remains the goal and noted customer impact requires pragmatic handling in the near term.

Outcome: No decision on making the headers optional. Directory submission checks will be implemented to verify server compliance.

4. Security Interest Group becomes a routing group

• @localden noted the group is dormant, and since security is a core pillar of the project a full shutdown would concern the community.
• @pcarleton proposed repurposing it as a routing group that does initial vetting of security issues and directs them to the appropriate working groups.
• Agreed to drop formal recurring meetings in favor of monthly digests and an open forum model driven by calls for agendas.

Outcome: Aligned. Security Interest Group converts to a routing group with an as-needed agenda and forum-based structure.

5. Transport working group lead: vote initiated

• @kurtisvg nominated a candidate as transport working group lead ahead of his own paternity leave in January, requesting formal approval from core maintainers.

Outcome: @kurtisvg to run a core-maintainer vote on the transport WG leads and secure sign-off from @dsp-ant and Den to finalize.

6. Agents, Triggers, and Extensions working groups consolidate

• @CaitieM20 proposed merging the Agents WG, the Triggers and Extensions WG, and the loosely affiliated Tasks work into a single group around early September, noting the Agents WG's focus had shifted toward agent-definition primitives that were unlikely to prove viable.
• @pja-ant agreed to close out the trigger specification for a vote before the merge.

Outcome: Aligned. Groups merge into a single entity to streamline alignment on messaging primitives. @CaitieM20 to coordinate timing with @clareliguori.

7. Skills-over-MCP spec updates: up for vote

• @pja-ant walked through key updates: clients must not materialize or download the entire list of skills and files ahead of time (protects servers from overload), and a skill's identity is defined by its server identifier plus URI rather than its name.
• @kurtisvg and @pja-ant discussed how skills link to other skills and resolved to review context mapping and URI handling.

Outcome: Proposal goes to a vote. The group to review the updated proposal and vote. @kurtisvg and @pja-ant to verify the name-to-URI mapping in the documentation and confirm the model correctly uses the URI for resource loading.

8. Review queue and proposal scheduling: discussion only

• @pcarleton said the DPoP feature is nearly ready for review, targeting the next one to two weeks. This prompted discussion of a formal review queue and submission forum.
• @kurtisvg and @pcarleton discussed Google Groups for core-maintainer permissions. @pcarleton noted the Google Workspace lacks the groups feature, so admin support from @localden or @dsp-ant is required.

Outcome: No formal decision. Queue and forum setup to continue.

Actions

• @localden: Move the recurring meeting to the MCP calendar.
• @CaitieM20: Start a thread with Peter and Clare on timing for merging the Agents, Triggers, and Tasks working groups.
• @CaitieM20: Review the updated Skills-over-MCP proposal documentation.
• The group: Review the updated Skills-over-MCP proposal and vote.
• @kurtisvg, @pja-ant: Verify the name-to-URI mapping in the documentation and whether the model correctly uses the URI for resource loading.
• @kurtisvg: Initiate the core-maintainer vote for transport WG leads and get sign-off from David and Den.