Skip to main content
lindvall
lindvall
Rookie
July 3, 2026
Question

Timeouts when connecting to MCP endpoint

  • July 3, 2026
  • 2 replies
  • 207 views

I’ve been running into problems with Claude reporting timeouts when I try to start a new session and use the Front MCP. 

 

In 8 of the last 12 sessions I’ve had exactly 60 seconds after a successful connect, the local log shows HTTP connection dropped after 60s uptime followed by Failed to open SSE stream: Gateway Time-out.

 

`initialize` on https://mcp.frontapp.com/mcp intermittently hangs with no response at all; observed 2026-06-20, 2026-07-01, and 2026-07-03 around 19:44–19:45 UTC today, with successful reconnect a minute later. Additionally the SSE stream (GET on the same endpoint) is consistently terminated with a Gateway Time-out after exactly 60 seconds of uptime, on essentially every session — looks like a gateway idle-timeout misconfiguration rather than a transient outage.

 

These issues can be resolved by reconnecting, but it really is a pain and makes it much less smooth to try to use MCP.

    2 replies

    Javier - Developer Relations
    Javier - Developer Relations
    Community Manager
    July 6, 2026

    Hi ​@lindvall, thanks for reporting this. I’ve created a ticket for the team to investigate

    Javier - Front Developer Relations
    eric.neto
    eric.neto
    Fronteer
    July 23, 2026

    Hi ​@lindvall, thank you for the detailed report.

    The 60-second drops were coming from the GET request on `/mcp`, which opens an SSE stream for server-initiated messages. Our MCP server runs stateless and never sends server-initiated messages, so that stream had nothing to push and just sat idle until the gateway's 60-second idle timeout closed it, surfacing as "HTTP connection dropped after 60s uptime" and "Failed to open SSE stream: Gateway Time-out." Because the stream never carried any data, those timeouts were effectively noise: your actual tool calls go over `POST /mcp` and were never affected by them.

    We've since changed `GET /mcp` to return `405 Method Not Allowed` (with an `Allow: POST` header), which the MCP Streamable HTTP spec explicitly permits for a server that offers no server-initiated messages. Compliant clients read that and stop opening the idle stream, so the 60-second Gateway Time-outs should no longer appear.

    Separately, the intermittent cases where `initialize` hangs with no response and then reconnects a minute later look like a distinct, transient issue rather than the idle timeout above. If you still see those after picking up the `GET /mcp` change, please send a few timestamps and we'll trace them.

    Please let us know if you're still seeing either the 60-second drops or the `initialize` hangs.

    Thank you!

    Terms & ConditionsAccessibility statement