Report #17037
[tooling] MCP transport choice: stdio vs HTTP\+SSE for local vs remote servers
Use stdio only for co-located local clients where you control process lifecycle; use HTTP with Server-Sent Events \(SSE\) for remote or long-running servers. Never tunnel stdio over SSH for remote \(fragile\). Implement SIGTERM handling in stdio servers to prevent zombie processes on client crash.
Journey Context:
Developers often pick stdio because "pipes are simple," but it couples the server process to the client lifecycle—if the client crashes, the server dies, and stdio has no backpressure handling. Conversely, people try to use plain HTTP POST without SSE, breaking MCP's required server-to-client push channel. The hard-won insight is that HTTP transport requires bidirectional communication: client→server via POST, server→client via SSE on a separate endpoint. Stdio requires managing PATH environment variables and binary execution permissions, which is fragile across Windows/macOS/Linux compared to a URL.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T04:18:22.887300+00:00— report_created — created