feat: proof of concept - MCP [not to be merged]

backend/openedx_ai_extensions/mcp/README.md

@@ -0,0 +1,135 @@
+# Model Context Protocol (MCP) - Proof of Concept
+
+This directory contains a **proof of concept (MVP)** demonstrating how Model Context Protocol (MCP) works. This is **NOT** the full OpenEdx AI Extensions integration yet, but rather a working example to understand MCP concepts and test the infrastructure.
+
+## Overview
+
+This MVP demonstrates the MCP architecture and workflow using a simple dice-rolling example. The implementation consists of three main components:
+
+1. **`server.py`** - A FastMCP server with an example tool (`roll_dice`)
+2. **`run_server.py`** - Script to run the MCP server in HTTP mode
+3. **`client_example.py`** - Example client showing how to interact with the MCP server
+
+**Note:** The actual OpenEdx-specific tools and integration will be implemented in future iterations. This MVP focuses on validating the MCP infrastructure and communication patterns.
+
+## Architecture
+
+The current implementation uses:
+- **FastMCP** - A framework for building MCP servers
+- **Streamable HTTP** transport - Allows the server to be exposed via HTTP
+- **LiteLLM** - For integrating the MCP server with language models
+
+### Example Tool: `roll_dice`
+
+The server currently implements a simple example tool that rolls dice. **This is a demonstration tool only** to show how MCP tools work. This pattern will later be extended to implement OpenEdx-specific operations like course management, user administration, content creation, etc.
+
+## Setup
+
+### Prerequisites
+
+Install the required dependencies:
+
+```bash
+pip install fastmcp litellm openai
+```
+
+### Environment Variables
+
+Set your OpenAI API key:
+
+```bash
+export OPENAI_API_KEY="your_openai_api_key_here"
+```
+
+Or update it directly in `client_example.py`.
+
+## Running the Server
+
+### Step 1: Start the MCP Server
+
+Run the server locally on port 9001:
+
+```bash
+python run_server.py
+```
+
+The server will start and listen on `http://127.0.0.1:9001/mcp`.
+
+### Step 2: Expose the Server with ngrok
+
+Since the MCP protocol requires a publicly accessible endpoint for certain use cases, you need to expose your local server using ngrok:
+
+```bash
+# Install ngrok if you haven't already
+# Visit https://ngrok.com/ to download and set up
+
+# Expose port 9001
+ngrok http 9001
+```
+
+ngrok will provide you with a public URL like:
+```
+https://abc123.ngrok-free.app
+```
+
+**Important**: Copy the ngrok URL (including the subdomain) as you'll need it for the client configuration.
+
+### Step 3: Update the Client Configuration
+
+Edit `client_example.py` and update the `server_url` with your ngrok URL:
+
+```python
+tools=[
+    {
+        "type": "mcp",
+        "server_label": "dice_server",
+        "server_url": "https://<your_ngrok_subdomain>.ngrok-free.app/mcp/",
+        "require_approval": "never",
+    },
+],
+```
+
+Replace `<your_ngrok_subdomain>` with your actual ngrok subdomain (e.g., `abc123.ngrok-free.app`).
+
+### Step 4: Run the Client Example
+
+In a new terminal (while the server and ngrok are still running):
+
+```bash
+python client_example.py
+```
+
+## Testing Workflow
+
+Here's the complete workflow for testing:
+
+1. **Terminal 1** - Start the MCP server:
+   ```bash
+   cd backend/openedx_ai_extensions/mcp
+   python run_server.py
+   ```
+
+2. **Terminal 2** - Expose with ngrok:
+   ```bash
+   ngrok http 9001
+   ```
+   Copy the ngrok URL from the output.
+
+3. **Terminal 3** - Run the client:
+   ```bash
+   # Update client_example.py with your ngrok URL first
+   python client_example.py
+   ```
+
+## Expected Output
+
+When running the client, you should see:
+
+1. List of available tools from the MCP server
+2. The AI model response after using the `roll_dice` tool
+
+Example:
+```
+Available resources: ['roll_dice']
+Response: <LiteLLM response object with dice roll results>
+```

backend/openedx_ai_extensions/mcp/__init__.py

@@ -0,0 +1,3 @@
+"""
+Model Context Protocol (MCP) integration for OpenEdx AI Extensions
+"""

backend/openedx_ai_extensions/mcp/client_example.py

@@ -0,0 +1,57 @@
+# client_litellm.py
+import asyncio
+from fastmcp import Client
+from litellm import responses
+import os
+import openai
+
+from fastmcp.client.transports import StreamableHttpTransport
+from fastmcp.server.auth.providers.bearer import RSAKeyPair
+from pydantic import SecretStr
+
+# Read key pair
+with open("private.pem", "r") as private_key_file:
+    private_key_content = private_key_file.read()
+
+with open("public.pem", "r") as public_key_file:
+    public_key_content = public_key_file.read()
+
+os.environ["OPENAI_API_KEY"] = "your_openai_api_key_here"
+
+
+openai_client = openai.OpenAI()
+key_pair = RSAKeyPair(
+    private_key=SecretStr(private_key_content),
+    public_key=public_key_content
+)
+
+async def main():
+    # Generate JWT token
+    token = key_pair.create_token(
+        subject="[email protected]",
+        issuer="https://<your_ngrok_subdomain>.ngrok-free.app",
+        audience="dice_server",
+        scopes=["read", "write"]
+    )
+
+
+    response = responses(
+        model="gpt-4.1-nano",
+        reasoning=None,
+        tools=[
+            {
+                "type": "mcp",
+                "server_label": "dice_server",
+                "server_url": "https://ce888b1c33bf.ngrok-free.app/mcp/",
+                "require_approval": "never",
+                "authorization": token,
+            },
+        ],
+        input="Roll a dice for me.",
+    )
+
+    print("Response:", response)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

backend/openedx_ai_extensions/mcp/private.pem

@@ -0,0 +1,28 @@
+-----BEGIN PRIVATE KEY-----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+-----END PRIVATE KEY-----

backend/openedx_ai_extensions/mcp/public.pem

@@ -0,0 +1,9 @@
+-----BEGIN PUBLIC KEY-----
+MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEA9VdHoi4doU03VDQXW/Af
+8OF9glcg8XkmF9vg6a+M0NTvBPsEGXKcNOfEulmN+A9g9MKtOt6VHldzHGsOYzPz
+dl/82HHOE6v9DngvpXhK3RZb0i4AaRUrM+NZ7WZoKcQIm/Okh9CfKBz2xtxQWbzu
+a4g+WBpfIK7ovoxrI5IKMAOQvYk12tyvEqcJNFT2hdrWEk6+32uIevcgFSWqvpkZ
+LEGU4afPVvZvq3Q3yZQpTg+elIH1o0yWEs7U7QeW2Gp+ZCBWsYnrW4CJ0rewEPD0
+W+PloMe4qMgaYi4pkLAM5qPO+aR+guBDHvPpigD9cc1lFH29Kw55tWwGhj/C90Dp
+9QIDAQAB
+-----END PUBLIC KEY-----

backend/openedx_ai_extensions/mcp/run_server.py

@@ -0,0 +1,10 @@
+#!/usr/bin/env python
+"""
+Run the FastMCP server in stdio mode
+"""
+from server import mcp
+
+
+if __name__ == "__main__":
+
+    mcp.run(transport="streamable-http")

backend/openedx_ai_extensions/mcp/server.py

@@ -0,0 +1,48 @@
+"""
+MCP Server implementation using FastMCP for OpenEdx AI Extensions
+
+This module provides a Model Context Protocol (MCP) server that exposes
+tools and resources for AI assistants to interact with the OpenEdx system.
+"""
+import logging
+from typing import Any
+from fastmcp import FastMCP, Context
+from starlette.responses import PlainTextResponse
+from starlette.requests import Request
+import random
+from fastmcp.server.auth.providers.jwt import StaticTokenVerifier
+from fastmcp.server.auth import BearerAuthProvider
+from fastmcp.server.auth.providers.jwt import JWTVerifier
+from fastmcp.server.dependencies import get_access_token, AccessToken
+from fastmcp.server.middleware.logging import LoggingMiddleware
+from fastmcp.server.middleware.error_handling import ErrorHandlingMiddleware
+
+logger = logging.getLogger(__name__)
+
+# Read public key for token validation
+with open("public.pem", "r") as public_key_file:
+    public_key_content = public_key_file.read()
+
+# Configure authentication provider
+auth = BearerAuthProvider(
+    public_key=public_key_content,
+    issuer="https://<YOUR_ISSUER>.ngrok-free.app",
+    audience="dice_server"
+)
+
+mcp = FastMCP(name="dice_server", port=9001, auth=auth)
+mcp.add_middleware(LoggingMiddleware())
+mcp.add_middleware(ErrorHandlingMiddleware(
+    include_traceback=True,
+    transform_errors=True,
+))
+
+
+@mcp.tool()
+def roll_dice(n_dice: int, context: Context) -> list[int]:
+    """Roll `n_dice` 6-sided dice and return the results."""
+    auth_user = get_access_token().client_id
+
+    if auth_user == "<YOUR_TEST_USER_ID>":
+        return [6 for _ in range(n_dice)]
+    return [random.randint(1, 6) for _ in range(n_dice)]