How to Add MCP Servers to ChatGPT

ChatGPT Joins the MCP Ecosystem

OpenAI added MCP support to ChatGPT, which means the same MCP servers that work with Claude and Cursor now work with ChatGPT too. Here's how to set it up.

What's Supported

ChatGPT's MCP support focuses on remote MCP servers — servers that run on the internet rather than on your local machine. This is different from Claude Desktop, which primarily runs MCP servers locally.

This means:

• Works: Remote/hosted MCP servers with HTTP endpoints
• Limited: Local MCP servers that run on your machine (these still need Claude Desktop or Cursor)

Setting Up Remote MCP Servers

In ChatGPT Settings

1. Open ChatGPT and go to Settings
2. Look for the Connections or MCP section
3. Click Add Connection
4. Enter the MCP server URL (e.g., https://mcp-server.example.com/mcp)
5. Authenticate if required
6. Save

The server's tools will now appear in your ChatGPT conversations.

Through GPT Builder

If you're building a custom GPT, you can add MCP servers in the GPT configuration:

1. Go to Explore GPTs > Create
2. In the Configure tab, find the Actions section
3. Add an MCP server endpoint
4. Configure authentication
5. Save and publish

Which MCP Servers Work with ChatGPT

Any MCP server that offers a remote endpoint works. Many popular servers are now available as hosted services:

• Brave Search — web search capabilities
• GitHub — repository management
• Slack — workspace communication
• Linear — project management
• Notion — workspace and document access

For local-only servers (filesystem, database connections), you'll need to either:
- Use them with Claude Desktop or Cursor instead
- Set up a hosted version behind your own endpoint
- Use a tunneling service to expose them

Testing Your Connection

After adding an MCP server, test it:

1. Start a new conversation in ChatGPT
2. Ask something that requires the server's capabilities
3. ChatGPT should indicate it's using the MCP tool
4. Verify the results are correct

Example: After connecting the GitHub MCP server, ask "List my recent pull requests" — ChatGPT should query GitHub directly.

Tips for ChatGPT + MCP

Combine with GPTs: Create a custom GPT that has specific MCP servers pre-configured for a workflow. A "Code Review GPT" with GitHub MCP, or a "Research GPT" with Brave Search MCP.

Check permissions carefully: When you connect an MCP server, you're giving ChatGPT access to that service. Review what permissions the server requests.

Use Projects: ChatGPT's Projects feature (similar to Claude's) lets you set context that persists across conversations within a project. Pair this with MCP servers for powerful workflows.

The Bigger Picture

MCP support in ChatGPT validates the protocol as the universal standard for AI tool integration. Tools built for one AI app now work across the ecosystem:

• Build an MCP server once
• Use it in Claude, ChatGPT, Cursor, VS Code, Windsurf
• Users aren't locked into any single AI vendor

This is good for everyone — tool builders reach a wider audience, and users get more choice.

Finding MCP Servers

Browse the full catalog of MCP servers on a-gnt.com. Each listing shows which AI apps are supported and includes setup instructions specific to your app. Filter by "ChatGPT compatible" to see servers with remote endpoints.

Whether you're a ChatGPT loyalist or use multiple AI apps, MCP servers expand what your AI can do — and now that expansion works everywhere.