AI Client Guide

How to Configure MCP Servers in ChatGPT

ChatGPT MCP configuration is covered here as a Model Context Protocol client guide page: what it is, how it works, when to use it, how to configure it safely, and how to verify the result. The guidance is grounded in the official MCP specification, SDK documen

MCP Support in ChatGPT

ChatGPT is a web-app built by OpenAI. It acts as an MCP host, embedding an MCP client that can connect to any MCP-compatible server. It supports stdio and streamable-http transport.

tools
resources
prompts

Prerequisites

  • ChatGPT installed (download)
  • Node.js 18+ or the runtime required by your target MCP server
  • An MCP server package or source code to configure
  • API credentials for your target integration

Configuration

{
  // .vscode/mcp.json
  "servers": {
    "example-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@mcp/example-server"]
    }
  }
}
  • command — the executable (npx, node, python, uvx)
  • args — arguments passed to the server process
  • env — environment variables injected at startup

Verifying the Connection

  1. Save the config file and fully restart ChatGPT.
  2. Look for the MCP server indicator — a hammer icon or server status badge in ChatGPT.
  3. Send a message asking the model to list available tools.
  4. Confirm all expected tools appear and a test call returns a valid response.

Troubleshooting

  • Server not visible → check logs, confirm the package name, restart ChatGPT.
  • Tools missing → verify capability negotiation; the server must expose a tools/list handler.
  • Auth failures → check env var names match exactly.

See MCP server not connecting and MCP tools not appearing

M
MCPServer.in Editorial

Editorial Team

Published: 2026-08-05
Updated: 2026-08-05

References & Technical Specifications

Frequently Asked Questions

Contextual information and technical support details regarding Model Context Protocol integration