@upstash/context7-mcp@4.0.1

neuml/txtai@upstash/context7-mcp@4.0.1Aug 11, 2026by github-actions[bot]

AI Summary

A patch release fixing a critical resource exhaustion issue in v4.0.0. By removing the forced SSE response mode, it prevents the exhaustion of gateway connection pools caused by unclosed SSE streams.

Key Highlights

Changed responseMode from forced 'sse' to SDK default 'auto'.
Prevented connection pool overflow (503 errors) by reducing concurrent upstream streams.
Modern-protocol responses are now plain JSON instead of streams.

Full Release Notes

### Patch Changes

- af7e4ad: Stop forcing `responseMode: "sse"` on the HTTP handler and use the SDK default `"auto"` instead. Forcing `"sse"` put every response on an SSE stream, and those streams were not released: concurrent upstream streams went from ~10 before v4.0.0 to over 5000, exhausting the gateway connection pool and returning 503 `reset reason: overflow` on `mcp.context7.com`. Traffic and latency were unchanged over that period, so the growth was not load.

  With `"auto"` a request is answered with a single JSON body unless a handler emits a related message before its result, which upgrades that one exchange to SSE. No tool emits progress today, so modern-protocol responses are now plain JSON. The 2025-era legacy fallback is constructed without a `responseMode` and still streams over SSE, so it is unaffected.