ktdreyer/openai-mcp

This is an OpenAI organization management MCP server that allows you to invite users to your OpenAI organization.

Tired of inviting users to your OpenAI org by hand? Tell your LLM to do it for you.

Example Queries

Once connected to an LLM, you can ask natural language questions like:

• "What is the status of my OpenAI invite? My email is user@example.com."
• "Please invite me to the OpenAI org. My email is user@example.com."

Security Note: This is a naive example where users provide their own email addresses. In a production deployment, you should pass authenticated user email addresses through a secure mechanism rather than trusting user input to the LLM.

LLM Instructions

Use this prompt when configuring your LLM to use this MCP server:

You are a helpful assistant who manages OpenAI organization invitations for our team.

Your capabilities:
- Check the status of pending invitations by email address
- Send new invitations with "reader" role access

Workflow:
- Always check if an invitation already exists before creating a new one
- If a user asks for an invite and already has one pending, tell them and provide the invite details
- If they need a new invite, you must have their email address - ask if they haven't provided it
- All invitations are for "reader" role access only

Be friendly and concise. If an operation fails, explain what went wrong clearly.

Prerequisites

Create two restricted OpenAI admin keys at https://platform.openai.com/settings/organization/admin-keys:

1. Read key with api.management.read scope
2. Write key with api.management.write scope

(This is an OpenAI limitation - a single restricted key cannot GET and POST the URL we need.)

Setup

Set both OpenAI Admin API keys as environment variables:

export OPENAI_ADMIN_KEY_READ="sk-admin-your-read-key-here"
export OPENAI_ADMIN_KEY_WRITE="sk-admin-your-write-key-here"

Optionally, set a default project ID to automatically add invited users as members:

export OPENAI_DEFAULT_PROJECT_ID="proj_abc123..."

Usage

For development with MCP inspector, run:

uv run mcp dev main.py

For use with an MCP client (eg 5ire), run:

uv run main.py

... and connect the client to a "remote" endpoint, http://127.0.0.1:8000/mcp.

Available Tools

(see main.py method docstrings for details)

• invite_reader(email) - Invite a user to your OpenAI organization as a reader (optionally adds them to the default project if OPENAI_DEFAULT_PROJECT_ID is set)
• retrieve_invite(email) - Retrieve an invite for a specific email address