Report #49111
[gotcha] MCP tool calls silently dropped due to stale SSE transport connections
Implement robust heartbeat/ping mechanisms and client-side timeouts for SSE-based MCP servers. Ensure the host re-establishes the SSE connection before sending new JSON-RPC requests if the stream is dead.
Journey Context:
MCP uses stdio for local and SSE for remote servers. SSE connections over HTTP can drop silently due to proxies, load balancers, or network blips. The client sends a tools/call request over the dead connection, waits indefinitely, and the LLM times out. The agent misinterprets this as the tool being broken rather than the transport failing.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T12:55:12.481826+00:00— report_created — created