Gatewaystack-chatgpt-starter: MCP Server Framework for AI 3rd-party Auth with OAuth/JWT

Gatewaystack Chatgpt Starter

GatewayStack is an MCP server framework that provides secure authentication and user data access for LLMs such as ChatGPT. It implements user authentication, permission control, and audit tracking through OAuth/JWT, solving the third-party authentication problem in AI applications.

Security Developer tools #AI Security Gateway #Authentication #User Data Access #MCP Server .TypeScript

rating : 2 points

downloads : 3.3K

update time : 2025-12-29

Open Site

What is the GatewayStack ChatGPT Secure Connector?

This is a Model Context Protocol (MCP) server specifically designed for ChatGPT, which solves the 'third-party authentication problem' in AI applications. When users access your internal tools (such as calendars, sales data, documents, etc.) through ChatGPT, this connector ensures that only the user themselves can access their own data, not others'. There is a security vulnerability in traditional AI applications: ChatGPT knows who the user is, but your backend system doesn't. This connector establishes a trust layer in the middle, securely transmitting the user's identity information to your system, ensuring the accuracy and security of data access.

How to use the GatewayStack Connector?

The usage process is very simple for end-users: 1. Users enable your tool in ChatGPT. 2. ChatGPT will automatically guide users to complete OAuth login (if required). 3. Users converse with ChatGPT as usual. 4. When ChatGPT needs to access user data, it will securely obtain it through the connector. For developers, you need to deploy this server, configure the OAuth provider, and integrate your tools.

Applicable Scenarios

This solution is particularly suitable for the following scenarios: • Enterprise employees access personal work data (calendars, emails, tasks) through ChatGPT. • Customer service AI needs to view specific customers' order histories. • Medical AI assistants need to access patients' medical records (only authorized doctors). • Sales teams analyze personal sales data through AI. • Any multi-tenant AI application that requires user authentication.

Main Features

OAuth Authentication Integration

Supports all major OAuth 2.0/OIDC providers (Auth0, Okta, Azure AD, Google, etc.), automatically handles the login process, ensuring the authenticity and reliability of user identities.

JWT Token Verification

Automatically verifies JWT tokens from ChatGPT, checks the issuer, audience, and signature, preventing forged requests.

Fine-grained Permission Control

Define the required permission scopes for each tool, ensuring that users can only access authorized functions.

Complete Audit Logs

Records all AI requests: who, when, to which model, and what operations were requested, meeting compliance requirements.

MCP Protocol Compatibility

Fully compatible with ChatGPT's Model Context Protocol, supporting all standard MCP operations: tool list, tool invocation, resource access, etc.

Flexible Deployment Options

Supports local development, cloud run (Cloud Run), and containerized deployment, providing complete CI/CD configuration examples.

Advantages

🔐 End-to-end security: Full authentication from user to backend, no security vulnerabilities.

👥 User data isolation: Ensure that each user can only access their own data and cannot see others' information.

🏢 Enterprise-ready: Supports SSO, permission management, and audit logs, meeting enterprise security requirements.

⚡ Plug-and-play: Provides complete infrastructure code for quick integration with existing tools.

🔧 Highly customizable: Supports custom identity mapping, tool definition, and permission logic.

📊 Complete observability: Built-in structured logs for easy monitoring and troubleshooting.

Limitations

📚 Learning curve: Requires understanding of concepts such as OAuth, JWT, and MCP for correct configuration.

🔧 Complex configuration: Requires correct setup of OAuth providers and permission scopes.

⚙️ Maintenance cost: Requires continuous server maintenance and dependency updates.

🔗 Dependence on external services: Requires available OAuth providers and a stable network connection.

🚀 Performance overhead: The additional authentication layer adds a small amount of latency.

💾 Resource requirements: Requires infrastructure to run the server.

How to Use

Environment Preparation

Clone the code repository, install dependencies, and copy the environment configuration file template.

Configure the OAuth Provider

Configure environment variables according to your identity provider (such as Auth0, Okta, etc.), including client ID, secret key, issuer URL, etc.

Define Your Tools

Add or modify tool definitions in the tools.ts file, including tool names, descriptions, required permissions, and implementation logic.

Local Testing

Start the development server, test if the tools work properly, and verify the authentication process.

Deploy to the Production Environment

Deploy the server using the provided Cloud Run configuration or your chosen platform, and configure the domain name and SSL certificate.

Configure in ChatGPT

Add the deployed server URL to ChatGPT's MCP server configuration, and users can start using it.

Usage Examples

Sales Data Query

Sales representatives query their personal sales performance through ChatGPT, and the system only returns the representative's data without revealing other colleagues' information.

Medical Record Access

Doctors view patients' medical records through an AI assistant. The system verifies the doctor's identity and permissions, ensuring that only authorized doctors can access specific patients' records.

Project Management

Project managers track team project progress through ChatGPT, and the system filters accessible projects and tasks based on user permissions.

Customer Service

Customer service representatives query customers' historical interaction records through an AI assistant to quickly understand customer issues and handling histories.

Frequently Asked Questions

Do I need programming knowledge to use this connector?

Which authentication providers are supported?

Is this solution secure?

How much does deployment cost?

What if ChatGPT cannot connect to my server?

Can I support multiple different backend systems simultaneously?

What's the difference between this solution and directly using ChatGPT plugins?

How long does it take to deploy and run?

Related Resources

Live Demo

Experience a full-featured demo directly in ChatGPT without deployment

Deployment Guide

Detailed deployment steps, including Cloud Run and CI/CD configuration

Auth0 Configuration Guide

Complete configuration instructions for using Auth0 as an identity provider

Other Identity Provider Guides

General guides for configuring providers such as Okta, Azure AD, Google, etc.

Troubleshooting Guide

Solutions to common problems and debugging techniques

GitHub Repository

Complete source code and latest updates

GatewayStack Architecture

Gain in-depth understanding of the underlying six-layer governance model and complete architecture

Detailed Explanation of the Third-Party Problem

Detailed explanation of the core security problem this solution aims to solve

🚀 Expose internal tools to LLMs — securely

This project provides a secure way to expose internal tools to LLMs. It allows users to access their data from inside ChatGPT, while also addressing the trust and governance issues between users, LLMs, and the backend.

🚀 Quick Start

git clone https://github.com/davidcrowe/gatewaystack-chatgpt-starter
cd gatewaystack-chatgpt-starter
npm install
cp .env.example .env
# Edit .env with your OAuth config
npm run dev

✨ Features

Trust and Governance: Creates a trust and governance layer between users, LLMs, and the backend, solving the three - party problem of user identity verification.
Secure User - Scoped Access: Enables users to securely access their data within ChatGPT, with cryptographic proof of user identity for each AI request.
OAuth and JWT Handling: Handles OAuth discovery, JWT verification, and per - tool scope enforcement, and maps the OAuth subject to the internal user ID.
End - to - End OAuth: Forwards the same OAuth access token downstream, ensuring end - to - end security.
Multiple Deployment Options: Supports local development, Cloud Run via Cloud Build, and any Node - compatible container runtime.
Customizable: Allows customization of tool registry, identity mapping, and backend execution.

📦 Installation

The installation steps are as follows:

git clone https://github.com/davidcrowe/gatewaystack-chatgpt-starter
cd gatewaystack-chatgpt-starter
npm install
cp .env.example .env
# Edit .env with your OAuth config
npm run dev

💻 Usage Examples

Basic Usage

The basic usage is to start the server as shown in the Quick Start section. Then, you can connect ChatGPT to the deployed MCP server to experience the full MCP + OAuth flow.

Advanced Usage

If you want to use real backend services:

Configure Proxyabl to call your backend endpoints.
Forward the OAuth token downstream.
Verify JWT in your backend and use sub/uid to scope data.

📚 Documentation

This repo is intended to be forked and adapted. It encodes a production - tested pattern for OAuth - protected MCP tools.

Live demo: docs/live-demo.md
Deploy your own instance: docs/deploy-your-own.md
Auth0 setup: docs/auth0-setup-guide.md
Other identity providers: docs/other-idps.md
Troubleshooting: troubleshooting.md
Backend integration pattern: docs/backend-integration-pattern.md
Env config guide: docs/env-config-guide.md

🔧 Technical Details

Trust and Governance Layer

Modern AI apps involve three actors: the user, the LLM, and the backend. There is no shared identity layer, creating data, policy, and audit gaps. This repo implements a solution to the three - party problem by adding a GatewayStack in the middle.

OAuth and JWT Handling

When ChatGPT calls the MCP server:

ChatGPT sends a request with a Bearer token to tools/list or tools/call.
The server verifies the JWT (JWKS / issuer / audience), enforces tool scopes, and maps sub to uid.
Tool execution can be local or forwarded to a backend using Proxyabl.

Deployment

This repo supports multiple deployment styles, including local development, Cloud Run via Cloud Build, and any Node - compatible container runtime.

📄 License

The project is licensed under the Apache - 2.0 license.

Additional Information

Endpoints

Property	Details
MCP	`POST /mcp`
OAuth discovery	`GET /.well-known/oauth-protected-resource`, `GET /.well-known/openid-configuration`
Debug	`GET /` (health), `GET /debug-token` (verifies the bearer token and shows `sub/aud/scope`)

Demo Tools

This starter ships with two tools:

whoami: Proves user identity is flowing (returns uid, sub, scopes).
echo: Echoes back a message.

Customization

You usually only need to change three things:

Tool registry: src/tools/tools.ts
Identity mapping: subjectToUid(sub, email?) in src/handlers/authHelpers.ts
Backend execution (optional): Configure Proxyabl to call your backend endpoints.

Common Issues

401 + invalid_token: issuer/audience mismatch, or JWKS not reachable.
ACCESS_TOKEN_IS_ENCRYPTED_JWE: Your auth provider is returning JWE; you need a JWS access token.
tools/list keeps prompting OAuth: ChatGPT isn't seeing a valid WWW - Authenticate header.

Repo Layout

src/gateway/toolGateway.ts: Main HTTP handler (MCP + well - known + REST tools).
src/handlers/*: Auth helpers, MCP handler, well - known handlers, misc helpers.
src/tools/tools.ts: Tool pack (scopes + descriptors + output shaping).
src/server/expressServer.ts: Local dev server.

⚠️ Important Note

This project is designed to establish the minimum viable security baseline for AI tools that handle real user data. Make sure to use JWS access tokens and ensure the issuer/audience and JWKS settings are correct.

💡 Usage Tip

If you are new to this project, start with the live demo to understand how it works. Then, you can fork the repo and customize it according to your needs.