Socket hang up error when using MCP server: fails on both `gpt5_generate` and `gpt5_messages` with `https://api.openai.com/v1/responses`

[beetz12]

Body:
I’ve encountered a persistent “socket hang up” error when using the MCP server with GPT-5:

• OS/Environment: (Mac OS X Sequoia 15.5, Node.js 24.5, Local)
• Server version: (@dannyboy2042/gpt5-mcp-server npm / git SHA if built from source)
• OpenAI API: GPT-5 access confirmed, key verified (works fine with direct curl calls)

Steps to Reproduce:

1. Register the MCP server in Claude Code using the recommended command, supplying my valid API key.
2. MCP server reports successful registration and shows as connected.
3. Initiate a search (e.g., text generation or conversation).
4. Result:
   First tries gpt5_generate → returns:

   https://api.openai.com/v1/responses failed, reason: socket hang up
   

   Falls back to gpt5_messages → same error.

I’ve attached a screenshot showing the error dialog (see below).

Additional Info:

• Direct curl requests to the same endpoint with the same key work perfectly.
• The issue occurs immediately after attempting to use the MCP server—but only with the MCP, not with direct OpenAI API use.
• Confirmed the API key and setup are correct in .env and config.
• No apparent network/firewall issues (other outbound requests function).
• MCP server confirms “connected” before making the request.

Screenshots:

Troubleshooting already attempted:

• Fresh API key
• Reinstalled/updated server
• Short and simple prompts for testing

Expected behavior:
MCP server should successfully relay the prompt to GPT-5 and return results, as happens when using curl directly.

Actual behavior:
Fails with “socket hang up” after successful connection and upon first use.

Any ideas or similar experiences? Would appreciate debugging and/or guidance.