4.3 KiB
2026-04-29 MCP: Transport Working Group (Weekly)
TL;DR
The meeting focused on the stateless transition timeline, SDK conformance
testing versioning, and proposed feature agendas for future releases. Discussion
centered on the transition of the initialization flow to serverdiscover under
standard IO and the introduction of an E-tag mechanism for optimistic
concurrency correctness. Tensions arose regarding whether standard IO can
natively support headers for features like E-tags, leading the group to identify
the need for exploring HTTP over standard IO. Proposals for streaming and raw
binary handling were identified as priorities for the release after next, while
encryption and compression proposals were tabled due to a lack of immediate
customer demand.
Detailed Summary
Stateless Migration & Conformance Testing
- Initialize Flow Removal: The transition to a stateless architecture
involves removing the initialization call in favor of a
serverdiscover(orinitialize server discover) call to determine protocol versions and remove session constraints. - Testing and Merging Timeline: The final lock-in date for SEPs is April 30th, and approved SEPs must have conformance tests merged by May 15th before they are finalized at the end of June. All Tier 1 SDKs (TypeScript and Python) are expected to implement changes by the release date.
- Protocol Versioning: Maintainers agreed on using a special version string,
DRAFT-2026-V1(represented as uppercase draft, dash, 2026, dash, lowercase V1), to denote the new protocol version during testing.
E-Tag Correctness Mechanism & Standard IO Challenges
- Correctness vs Caching: Geoff Goodman proposed an E-tag mechanism (or fencing token) where a tools list provides identifiers, and the client supplies that identifier on tool calls. This allows the server to verify if the LLM is using valid tool definitions.
- Tension Point - Opaque Strings and State: Andreas Schlapbach and Stephen Halter raised concerns that returning opaque strings could act as a backdoor to stateful behavior. However, Geoff Goodman and Stephen Halter favored server-defined opaque strings for simplicity.
- Tension Point - Standard IO Limitations: Standard IO lacks a native
headers mechanism to convey E-tags, leading to a discussion on putting
metadata in
_meta. This prompted core maintainers to suggest investigating "HTTP over standard IO" to unify standard IO and HTTP features.
Future Protocol Feature Agendas
- Streaming & Binary Handling: Peter Alexander proposed streaming results and raw binary handling for large payloads as priorities for future releases, intersecting with the file upload working group.
- Compression & Encryption: Andreas Schlapbach proposed supporting message compression (e.g., GZip, Brotli) and message signing/encryption to guarantee message integrity in regulated environments.
- Tension Point - Encryption Utility vs Observability: Geoff Goodman and Stephen Halter questioned the utility of encryption over TLS, arguing it adds payload opacity and a loss of observability. Kurtis Van Gent emphasized that without active client demand, encryption should not be a priority.
Next Steps & Timeline
Core maintainers and SDK developers will finalize conformance tests for the stateless protocol by the May 15th deadline. Geoff Goodman will draft an SEP outlining the E-tags mechanism focusing on optimistic concurrency control. Andreas Schlapbach will contact Saxo Bank to determine if highly regulated financial institutions have verified use cases for payload encryption.
Action Items
- Andreas Schlapbach Post comments regarding the stateless SEP document, focusing on the initialization requirements.
- Stephen Halter Implement the protocol version string
DRAFT-2026-V1in the SDK and ensure conformance tests recognize it. - Geoff Goodman Draft a new SEP outlining the E-tags mechanism for optimistic concurrency control.
- Andreas Schlapbach Talk to Saxo Bank contacts regarding encryption and compression messaging needs and return with example user journeys.
Source: Meeting Notes — April 29, 2026