MCP OAuth2.1 with AWS Cognito Example

MCP OAuth2.1 with AWS Cognito Example

Demonstrates how to secure an MCP server with OAuth 2.1 using AWS Cognito, with support for dynamic client registration and client ID metadata documents.

Category
Visit Server

README

MCP + OAuth2.1 + AWS Cognito Example

Overview

This repository demonstrates how to secure a Model Context Protocol (MCP) server using OAuth 2.1 authorization flows, implemented entirely with Node.js and Express.js. While this example uses AWS Cognito as the backing authorization server, the implementation is provider-agnostic and can work with any OAuth 2.1 compliant authorization server.

Based on the MCP Authorization Specification (version 2025-11-25), this project showcases:

  • MCP server acting as a Resource Server (RS) with generic OAuth endpoints
  • Provider-agnostic OAuth 2.1 implementation (example uses AWS Cognito)
  • OAuth 2.1 Authorization Code Flow with PKCE and RFC 8707 Resource Indicators
  • Protected Resource Metadata (PRM) document discovery
  • Fully dynamic authorization server metadata discovery
  • Dynamic Client Registration (DCR) support
  • Client ID Metadata Documents (CIMD) support
  • Enhanced security features from MCP 2025-11-25 specification
  • Three client implementations:
    • Static client with pre-configured credentials
    • Auto-discovery client with dynamic registration (DCR)
    • Metadata client using Client ID Metadata Documents (CIMD)

Provider-Agnostic Design

This implementation follows OAuth 2.1 standards to ensure compatibility with any compliant authorization server:

  • MCP Server: Exposes standard OAuth metadata endpoints and proxies to the backing authorization server
  • Clients: Discover authorization servers dynamically without hardcoded provider-specific logic
  • Token Validation: Uses discovered JWKS URIs and issuer information from authorization server metadata
  • Flexible Backend: While Cognito is used as an example, any OAuth 2.1 server can be substituted

Understanding the New MCP Authorization Spec

The new MCP Authorization Specification introduces a clean separation between Resource Servers and Authorization Servers, making it easier to integrate with existing identity providers like AWS Cognito, Okta, Auth0, and others.

Key components of the specification:

  1. Protected Resource Metadata (PRM) document

    • The MCP server serves this document at /.well-known/oauth-protected-resource
    • Contains information about authorization servers, supported scopes, etc.
    • Follows RFC9728 (OAuth 2.0 Protected Resource Metadata)
  2. Discovery Process

    • When a client receives a 401 Unauthorized response, the WWW-Authenticate header contains a pointer to the PRM document
    • Client fetches the PRM document to discover the authorization server URL
    • Client fetches authorization server metadata dynamically from the discovered URL (no hardcoded endpoints)
  3. OAuth 2.1 Authorization

    • Authorization Code flow with PKCE
    • Bearer token usage for authenticated requests
    • Dynamic token validation using discovered JWKS URIs and issuer information
  4. Client Registration Priority (MCP 2025-11-25)

    • Pre-registered client credentials (if available for the server)
    • Client ID Metadata Documents (if authorization server advertises client_id_metadata_document_supported: true)
    • Dynamic Client Registration (RFC7591 fallback)
    • Prompt user for manual configuration (last resort)
  5. Dynamic Client Registration (DCR)

    • Allows clients to automatically register with new MCP servers
    • Eliminates the need for manual client registration processes
    • Follows RFC7591 (OAuth 2.0 Dynamic Client Registration Protocol)
  6. Client ID Metadata Documents (CIMD)

    • Client publishes its OAuth metadata at an HTTPS URL
    • The URL itself becomes the client_id
    • Authorization server fetches and validates the metadata document
    • No pre-registration or database storage required
    • Follows draft-ietf-oauth-client-id-metadata-document

This implementation showcases how to apply these concepts in a provider-agnostic way. The example uses AWS Cognito with custom Dynamic Client Registration through API Gateway endpoints and Lambda functions, but the core OAuth flow works with any compliant authorization server.

Architecture

Client → MCP Server → Authorization Server (e.g., AWS Cognito)
        (Resource Server)    (OAuth 2.1 Provider)
  1. Client sends a request without a token.
  2. MCP server responds with 401 Unauthorized + WWW-Authenticate header pointing to PRM metadata.
  3. Client retrieves PRM, discovers the Authorization Server URL dynamically.
  4. Client fetches authorization server metadata and performs OAuth 2.1 Authorization Code flow (with PKCE).
  5. Client obtains an access token and retries request to MCP server.
  6. MCP server validates token using dynamically discovered JWKS and grants access to the protected resource.

For detailed overview, see the Architecture Overview.

Diagrams:

Dynamic Client Registration (DCR)

This implementation includes support for OAuth 2.1 Dynamic Client Registration, allowing clients to:

  1. Dynamically discover the MCP server and authorization endpoints
  2. Register themselves with the authorization server
  3. Obtain credentials for the OAuth flow

The DCR flow works as follows:

  1. Client discovers the MCP server's protected resource metadata
  2. Client discovers the authorization server (Cognito)
  3. Client registers with the DCR endpoint in API Gateway
  4. Registration creates a Cognito app client and returns credentials
  5. Client uses these credentials for the standard OAuth 2.1 flow

Implementation Note: AWS Cognito does not natively support Dynamic Client Registration as specified in OAuth 2.0 DCR (RFC7591). This implementation bridges this gap by using:

  • API Gateway endpoints to provide the DCR API interface
  • Lambda functions to create Cognito app clients programmatically
  • DynamoDB to store the registration data

This approach allows us to maintain compliance with the MCP specification's DCR recommendation while leveraging AWS Cognito for robust authentication and authorization.

Security Note: This implementation uses anonymous DCR without additional authentication. For production environments, consider adding:

  • Rate limiting on registration requests
  • Client authentication (mTLS, initial access tokens)
  • Approval workflow for new clients
  • Limited scope access for dynamically registered clients

See our DCR Security Recommendations to enhance the security of the registration process.

Client ID Metadata Documents (CIMD)

The MCP Authorization Specification (2025-11-25) introduced Client ID Metadata Documents as the recommended client registration method. CIMD allows a client to use an HTTPS URL as its client_id, with the URL hosting a JSON document describing the client's OAuth metadata.

How CIMD Works

  1. Client publishes a metadata document at a URL (e.g., http://localhost:3003/client-metadata.json)
  2. During OAuth authorization, the client uses this URL as its client_id
  3. The authorization server fetches the metadata document from the URL
  4. The server validates the document structure and redirect URIs
  5. The client proceeds with a standard OAuth 2.1 flow

CIMD Metadata Document Example

{
  "client_id": "http://localhost:3003/client-metadata.json",
  "redirect_uris": ["http://localhost:3003/callback"],
  "client_name": "MCP CIMD Demo Client",
  "grant_types": ["authorization_code"],
  "response_types": ["code"],
  "token_endpoint_auth_method": "none"
}

CIMD Authorization Proxy

AWS Cognito does not natively support Client ID Metadata Documents, just as it doesn't natively support DCR. This implementation bridges CIMD transparently through the MCP server's authorization proxy:

  • The MCP server advertises client_id_metadata_document_supported: true in authorization server metadata
  • The MCP server overrides authorization_endpoint and token_endpoint in metadata to point to itself
  • When a client presents a URL-based client_id:
    1. The proxy fetches and validates the client's metadata document
    2. Transparently creates a Cognito app client via the existing DCR infrastructure
    3. Forwards the request to Cognito with the mapped credentials
  • When a client presents a standard client_id (pre-registered or DCR): requests are passed through to Cognito unchanged

The metadata-client has zero custom code for CIMD — it simply uses its metadata URL as client_id with standard OAuth endpoints, just like any spec-compliant client. All bridging is handled server-side.

CIMD Security

The authorization proxy includes:

  • SSRF protection (blocks private IP ranges, enforces HTTPS in production)
  • Strict client_id validation (must match the metadata URL exactly)
  • Redirect URI validation
  • Response size limits (64KB) and fetch timeouts (5s)
  • In-memory caching to prevent redundant fetches

Note: In development, http://localhost URLs are permitted. Production deployments must use HTTPS.

Quick Start

Prerequisites

  • Node.js 18+ installed
  • AWS test account with access to:
    • Cognito for Authorization Server (1 user pool, 2 app clients)
    • API Gateway / Lambda / DynamoDB for DCR and CIMD bridge (2 resources, 2 functions, 1 table)
    • CloudFormation for deploy (1 stack)
  • Basic knowledge of OAuth 2.1 flows

Setup

  1. Clone the repository

    git clone https://github.com/empires-security/mcp-oauth2-aws-cognito.git
    cd mcp-oauth2-aws-cognito
    
  2. Install dependencies for clients and server

    npm run install:all
    
  3. Deploy AWS resources

    npm run deploy
    
  4. Review generated .env files in:

    • src/client/.env
    • src/auto-client/.env
    • src/metadata-client/.env
    • src/mcp-server/.env
    • Compare with .env.example files
    • Manually verify/update CLIENT_SECRET if needed

Running the Application

  1. Start all services (server + 3 clients)

    npm run dev
    
  2. Visit http://localhost:3000 to test the pre-registered client OAuth flow

  3. Sign Up for a New User

    • Click the "Log in" button
    • Select "Sign up" in the Cognito hosted UI
    • Create a new user account
    • Verify your account by entering the confirmation code sent to your email
    • After successful verification, you'll be redirected back to the application
  4. Click the "Fetch MCP Data" button to make an authenticated request to the MCP server

  5. Visit http://localhost:3002 to test the DCR flow (auto-discovery client with Dynamic Client Registration)

  6. Visit http://localhost:3003 to test the CIMD flow (Client ID Metadata Document client)

Cleanup

  1. Cleanup AWS resources
    npm run cleanup
    

For detailed setup instructions, see the Setup Guide.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

  1. Fork the repository
  2. Create your feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes (git commit -m 'Add some amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

References

License

This project is licensed under the MIT License - see the LICENSE file for details.

Authors

  • Empires Security Labs 🚀

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured