Skip to content

E2E Tests Core (proxy) broken on all PRs: batch request now 400s in TimeStreamableHttpMcpServer #5932

Description

@samuv

Summary

The E2E Tests Core (proxy) shard is failing deterministically on every open PR, including PRs that don't touch proxy/transport code at all:

Failing spec: TimeStreamableHttpMcpServer > when starting the time server with streamable-http proxy > should respond to a single get_current_time request and a batch request (test/e2e/stdio_proxy_over_streamable_http_mcp_server_test.go:42)

Assertion: the batch request part now returns HTTP 400 where 200 is expected.

Likely cause

This looks like fallout from the recent stateless-proxy work on main (e.g. #5884, "Serve MCP 2026-07-28 stateless requests through the transparent proxy") — the MCP 2026-07-28 revision removed JSON-RPC batching, so the proxy rejecting a batch with 400 may be intentional spec-conformant behavior with a test that wasn't updated, or an unintended regression for older-protocol clients.

Either way the spec and the proxy currently disagree, and every PR's required check is red because of it.

Impact

All open PRs show mergeState=UNSTABLE on this shard regardless of their content. Example failing job: https://github.com/stacklok/toolhive/actions/runs/29992236950/job/89174538909

Metadata

Metadata

Assignees

Labels

needs-triageIssue needs initial triage by a maintainer

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions