MCP & Integrations

📖 3 min read openaimcpintegrationchatgptconnectors

OpenAI's MCP support — Model Context Protocol for connecting AI to external tools. MCP Apps in ChatGPT, Secure MCP Tunnel, Connectors, ChatGPT Actions, and building custom integrations.

Key Takeaways

OpenAI supports MCP across Codex, ChatGPT, and the Agents SDK — same protocol as Anthropic
MCP Apps in ChatGPT: Build interactive apps that run inside ChatGPT's interface using MCP
ChatGPT Connectors: Connect to Google Drive, Slack, Notion, and 200+ services
Secure MCP Tunnel: End-to-end encrypted connection for remote MCP servers

OpenAI supports the Model Context Protocol across its entire platform. The same MCP servers work with Codex, ChatGPT, the Agents SDK, and any MCP-compatible client.

MCP in Codex

Codex reads .codex/mcp.json and automatically connects to configured servers:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "${GITHUB_TOKEN}" }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"]
    },
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"]
    }
  }
}

codex "find all open PRs that need review, check the code quality, and comment on each"

MCP Apps in ChatGPT

The ChatGPT Apps SDK lets you build interactive MCP-powered apps that run inside ChatGPT’s interface:

# Build a restaurant reservation app as an MCP server
from chatgpt_apps import ChatGPTApp

app = ChatGPTApp("restaurant-reservation")

@app.tool()
def find_tables(restaurant: str, date: str, party_size: int):
    """Find available tables at a restaurant."""
    # Your reservation system logic
    return {"tables": [{"time": "7:00 PM", "seats": 4}]}

@app.tool()
def book_table(restaurant: str, date: str, time: str, party_size: int):
    """Book a table."""
    return {"confirmation": "CONF-12345"}

Users can then say: “Book a table for 4 at Per Se tomorrow at 7 PM” — and your MCP app handles it inside ChatGPT.

ChatGPT Connectors

Connect ChatGPT to your tools and data without writing code:

Connector	What It Enables
Google Drive	Search and read documents
Slack	Search channels, send messages
Notion	Query and update pages
GitHub	Browse repos, manage issues
Jira	Create and update tickets
Linear	Manage project issues

Secure MCP Tunnel

For remote/cloud MCP servers, OpenAI provides end-to-end encryption:

# Start a secure tunnel from your local MCP server to ChatGPT
openai mcp tunnel \
  --server "npx -y @modelcontextprotocol/server-postgres" \
  --name "production-db"

This enables ChatGPT to connect to databases and services behind your firewall without exposing them to the internet.

ChatGPT Actions

Actions let you add custom capabilities to ChatGPT without building a full MCP app:

{
  "action": {
    "name": "get_stock_price",
    "description": "Get current stock price by ticker symbol",
    "endpoint": "https://your-api.com/stock-price",
    "parameters": {
      "ticker": { "type": "string", "description": "Stock ticker symbol" }
    }
  }
}

Ecosystem Comparison

Protocol	What It Does	Supported By
MCP	Connect AI to external tools and data	OpenAI, Anthropic, VS Code, Cursor, 32+ tools
ChatGPT Actions	Add custom capabilities to ChatGPT	ChatGPT
ChatGPT Apps SDK	Build interactive MCP apps in ChatGPT	ChatGPT
Function Calling	Model-triggered API calls	All OpenAI API models
SKILL.md	Teach agents repeatable workflows	Codex, Claude Code, Cursor, Copilot

For the MCP protocol in general (architecture, building servers, security), see the Claude MCP page.