[MuhammadFaizan-prog]

v1.x port of #2662. Refs #2661.

Problem

Two tests in test/server/streamableHttp.test.ts call reader.read() once and then assert on more than one SSE event:

• should handle batch request messages with SSE stream for responses: expects both the req-1 and req-2 responses in the first chunk.
• should store and replay MCP server tool notifications: expects the replayed notification in the first chunk after reconnecting with Last-Event-ID.

fetch gives no guarantee that separately written SSE events share a chunk. On Node 26.7.0 they arrive in separate reads, so on v1.x as it is:

Tests  4 failed | 183 passed (187)
AssertionError: expected 'event: message\ndata: {"result":{"tools"…"id":"req-1"}' to contain '"id":"req-2"'
AssertionError: expected 'event: message\nid: …\ndata: {}' to contain 'Second notification from MCP server'

The transport behaves correctly; only the tests assume coalescing. The same file passes on Node 22.

Change (test-only)

Adds the same readSSEUntil(response, predicate, timeoutMs = 2000) helper as #2662: it accumulates decoded chunks until the predicate holds, the stream ends, or the timeout cancels the reader, then releases the lock. It is used at the three single-read sites in the two tests. The replay test now cancels through sseResponse.body.cancel() because the helper has released the reader. No production code changes, so no changeset.

I kept this to the two failing tests. The extra cleanup in #2662's follow-up commit (folding the other inline read loops into the helper) is left out to keep the v1.x diff minimal.

Testing

• Node 26.7.0: before, 4 failed / 183 passed; after, 187 / 187 (repeated runs).
• Node 22: 187 / 187 before and after.
• Sanity check: with a predicate that can never match, the test fails after the 2 s bound instead of hanging.
• npm run typecheck, Prettier and ESLint are clean on the changed file.

Credit

The failure report, repro and the read-until-predicate approach come from @jstar0 (#2661). The helper is the one from #2662. @Parithosh-Varma and @shi-felix also prepared ports for main.

AI disclosure

I used an AI assistant (Hark) to reproduce the failure on Node 26 and to draft this patch and description. I reviewed every line, ran the tests myself, and can explain the change.

• Test-only; no public API or behaviour change
• Fails before / passes after on Node 26.7.0
• typecheck / prettier / eslint clean

[changeset-bot]

⚠️ No Changeset found

Latest commit: fc4b380

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@modelcontextprotocol/sdk@3174

commit: fc4b380