MCP OAuth Gateway
A transparent proxy server that simplifies authentication by chaining its own OAuth layer with an upstream MCP server's credentials. It manages dual token sets behind a single interface, enabling secure and streamlined access to protected MCP resources.
README
MCP OAuth Gateway
Quick Start
Prerequisites
- Python 3.10+
- uv (recommended) or pip
Create a GitHub OAuth App
- Go to GitHub Developer Settings → OAuth Apps → New OAuth App
- Set Authorization callback URL to
http://localhost:8002/upstream/callback - Fill in any Application name and Homepage URL
- Click Register application
- Copy the Client ID and generate a Client Secret
Install & Run
# Clone and install
git clone https://github.com/abj453demo/mcp-oauth-gateway.git
cd mcp-oauth-gateway
uv venv && source .venv/bin/activate
uv pip install -e .
# Start the gateway (proxies to GitHub's remote MCP server)
mcp-oauth-gateway --port=8002 \
--upstream-rs=https://api.githubcopilot.com/mcp/ \
--upstream-client-id=<YOUR_GITHUB_CLIENT_ID> \
--upstream-client-secret=<YOUR_GITHUB_CLIENT_SECRET> \
--upstream-authorize-endpoint=https://github.com/login/oauth/authorize \
--upstream-token-endpoint=https://github.com/login/oauth/access_token
Replace <YOUR_GITHUB_CLIENT_ID> and <YOUR_GITHUB_CLIENT_SECRET> with the values from your GitHub OAuth App.
The gateway will be available at http://localhost:8002. Point your MCP client at http://localhost:8002/mcp.
Gateway Credentials
When prompted at the gateway login screen (Screen 1):
- Username:
gateway_user - Password:
gateway_pass
Configurable via MCP_GATEWAY_USERNAME and MCP_GATEWAY_PASSWORD environment variables.
After gateway login, you'll be redirected to GitHub for OAuth authorization (Screen 2).
Overview
The MCP OAuth Gateway is a transparent proxy that sits between an MCP client and an upstream MCP server. It implements its own OAuth 2.1 layer and chains it with the upstream's OAuth, so the client sees a single auth surface while two independent token sets are managed behind the scenes.
The gateway acts as both an Authorization Server (AS) and a Resource Server (RS) to the client. To the upstream, it acts as a regular OAuth client.
┌──────────┐ ┌─────────────────────┐ ┌─────────────────────────────┐
│ Client │──────▶│ Gateway (AS + RS) │──────▶│ GitHub OAuth + MCP Server │
│ (Cascade) │◀──────│ localhost:8002 │◀──────│ api.githubcopilot.com/mcp/ │
└──────────┘ └─────────────────────┘ └─────────────────────────────┘
Client Registration
The gateway supports OAuth 2.0 Dynamic Client Registration (RFC 7591).
- Client discovers the gateway via
GET /.well-known/oauth-protected-resource, which returns the gateway as both the resource and its own authorization server. - Client fetches
GET /.well-known/oauth-authorization-serverto learn the gateway's OAuth endpoints (/authorize,/token,/register). - Client calls
POST /registerwith its redirect URIs and grant types. The gateway stores the client in memory and returns aclient_idandclient_secret.
The gateway uses pre-configured credentials (--upstream-client-id / --upstream-client-secret) to authenticate with the upstream AS (e.g., a GitHub OAuth App). For upstreams that support it, dynamic registration is also available.
Two-Screen Auth Flow
The authorization flow chains two OAuth flows into one client-facing redirect sequence.
Client Gateway Upstream AS
│ │ │
├─ GET /authorize ─────────▶│ │
│ ├─ redirect to /login │
│◀──────────────────────────┤ (Screen 1: gateway creds)│
│ │ │
├─ POST /login/callback ───▶│ │
│ (gateway_user/pass) ├─ redirect to upstream ───▶│
│ │ /authorize (Screen 2) │
│◀──────────────────────────┤◀──────────────────────────┤
│ │ │
│ (user logs in upstream) │ │
│───────────────────────────┼──▶ upstream callback ────▶│
│ │◀── upstream code ─────────┤
│ │ │
│ ├─ exchange upstream code │
│ │ for upstream tokens ─────▶
│ │◀── upstream access_token ─┤
│ │ + refresh_token │
│ │ │
│◀─ redirect with gw code ──┤ │
│ │ │
├─ POST /token (gw code) ──▶│ │
│◀── concatenated tokens ───┤ │
Step-by-step
- Client →
GET /authorize— Gateway stores the client's redirect URI, PKCEcode_challenge, and state. Redirects to its own/loginpage. - Screen 1: Gateway login — User enters gateway credentials. On success, the gateway generates a PKCE pair for the upstream and redirects the user to the upstream AS's
/authorize. - Screen 2: Upstream login — User authenticates with the upstream. The upstream AS redirects back to the gateway's
/upstream/callbackwith an auth code. - Gateway callback — The gateway exchanges the upstream auth code for upstream access + refresh tokens (using PKCE). It generates a gateway auth code, stashes the upstream tokens, and redirects back to the client's original
redirect_uriwith the gateway auth code. - Client →
POST /token— Client exchanges the gateway auth code (with its own PKCE verifier). The gateway creates its own access + refresh tokens, pairs them with the upstream tokens, and returns concatenated tokens to the client.
Token Format
Tokens returned to the client are base64-encoded pairs:
access_token = base64url( gateway_access_token + ":" + upstream_access_token )
refresh_token = base64url( gateway_refresh_token + ":" + upstream_refresh_token )
The client treats these as opaque strings. The gateway splits them on every request to validate the gateway half and forward the upstream half.
expires_in is set to min(gateway_ttl, upstream_ttl) so the client refreshes before either token expires.
MCP Proxying
On every POST /mcp request:
- Extract the
Bearertoken from theAuthorizationheader. - Look up the token record — validate the gateway access token (expiry, revocation).
- Extract the upstream access token from the pair.
- Forward the request to the upstream RS with
Authorization: Bearer <upstream_token>, passing throughContent-Type,Accept,Mcp-Session-Id, andMcp-Protocol-Versionheaders. - Stream the upstream response (including SSE) back to the client.
GET (SSE streams) and DELETE (session teardown) are proxied similarly.
Token Refresh
The client only interacts with the gateway for refresh — it never contacts the upstream directly.
- Client sends
POST /tokenwithgrant_type=refresh_tokenand the concatenated refresh token. - Gateway splits the refresh token, validates the gateway half.
- If a real upstream refresh token exists, the gateway calls the upstream AS's
/tokenwithgrant_type=refresh_token. If the upstream doesn't use refresh tokens (e.g., GitHub), the existing upstream access token is reused. - Gateway revokes old tokens, creates new gateway + upstream pairs, and returns new concatenated tokens.
In-Memory State
All state is held in memory (no database). Key stores:
| Store | Contents | Lifetime |
|---|---|---|
clients |
Registered OAuth clients | Until restart |
auth_codes |
Gateway auth codes | Consumed on /token exchange |
tokens / refresh_tokens |
Gateway-issued tokens | Until expiry or revocation |
state_mapping |
In-flight authorize flow state | Consumed on callback |
upstream_tokens |
Raw upstream token responses | Consumed on /token exchange |
GatewayTokenStore |
Concatenated token → token pair records | Until expiry or revocation |
Configuration
| Flag / Env Var | Purpose |
|---|---|
--upstream-rs |
Upstream MCP Resource Server URL (required) |
--upstream-as |
Upstream AS URL (auto-discovered from RS if omitted) |
--upstream-client-id / MCP_GATEWAY_UPSTREAM_CLIENT_ID |
Pre-configured upstream client ID (skips dynamic registration) |
--upstream-client-secret / MCP_GATEWAY_UPSTREAM_CLIENT_SECRET |
Pre-configured upstream client secret |
--upstream-authorize-endpoint / MCP_GATEWAY_UPSTREAM_AUTHORIZE_ENDPOINT |
Direct upstream authorize URL (skips .well-known discovery) |
--upstream-token-endpoint / MCP_GATEWAY_UPSTREAM_TOKEN_ENDPOINT |
Direct upstream token URL (skips .well-known discovery) |
MCP_GATEWAY_USERNAME / MCP_GATEWAY_PASSWORD |
Gateway login credentials (default: gateway_user / gateway_pass) |
Recommended Servers
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.