Fledge MCP Server

This is a Model Context Protocol (MCP) server that connects Fledge functionality to Cursor AI, allowing the AI to interact with Fledge instances via natural language commands.

Prerequisites

• Fledge installed locally or accessible via API (default: http://localhost:8081)
• Cursor AI installed
• Python 3.8+

Installation

1. Clone this repository:

git clone https://github.com/Krupalp525/fledge-mcp.git
cd fledge-mcp

2. Install the dependencies:

pip install -r requirements.txt

Running the Server

1. Make sure Fledge is running:

fledge start

2. Start the MCP server:

python mcp_server.py

For secure operation with API key authentication:

python secure_mcp_server.py

3. Verify it's working by accessing the health endpoint:

curl http://localhost:8082/health

You should receive "Fledge MCP Server is running" as the response.

Connecting to Cursor

1. In Cursor, go to Settings > MCP Servers

2. Add a new server:

3. For the secure server, configure the "X-API-Key" header with the value from the api_key.txt file that is generated when the secure server starts.

4. Test it: Open Cursor's Composer (Ctrl+I), type "Check if Fledge API is reachable," and the AI should call the `validate_api_connection` tool.

Available Tools

Data Access and Management

1. get_sensor_data: Fetch sensor data from Fledge with optional filtering by time range and limit

2. list_sensors: List all sensors available in Fledge

3. ingest_test_data: Ingest test data into Fledge, with optional batch count

Service Control

4. get_service_status: Get the status of all Fledge services

5. start_stop_service: Start or stop a Fledge service by type

6. update_config: Update Fledge configuration parameters

Frontend Code Generation

7. generate_ui_component: Generate React components for Fledge data visualization

8. fetch_sample_frontend: Get sample frontend templates for different frameworks

9. suggest_ui_improvements: Get AI-powered suggestions for improving UI code

Real-Time Data Streaming

10. subscribe_to_sensor: Set up a subscription to sensor data updates

11. get_latest_reading: Get the most recent reading from a specific sensor

Debugging and Validation

12. validate_api_connection: Check if the Fledge API is reachable

13. simulate_frontend_request: Test API requests with different methods and payloads

Documentation and Schema

14. get_api_schema: Get information about available Fledge API endpoints

15. list_plugins: List available Fledge plugins

Advanced AI-Assisted Features

16. generate_mock_data: Generate realistic mock sensor data for testing

Testing the API

You can test the server using the included test scripts:

# For standard server
python test_mcp.py

# For secure server with API key
python test_secure_mcp.py

Security Options

The secure server (secure_mcp_server.py) adds API key authentication:

1. On first run, it generates an API key stored in api_key.txt

2. All requests must include this key in the X-API-Key header

3. Health check endpoint remains accessible without authentication

Example API Requests

# Validate API connection
curl -X POST -H "Content-Type: application/json" -d '{"name": "validate_api_connection"}' http://localhost:8082/tools

# Generate mock data
curl -X POST -H "Content-Type: application/json" -d '{"name": "generate_mock_data", "parameters": {"sensor_id": "temp1", "count": 5}}' http://localhost:8082/tools

# Generate React chart component
curl -X POST -H "Content-Type: application/json" -d '{"name": "generate_ui_component", "parameters": {"component_type": "chart", "sensor_id": "temp1"}}' http://localhost:8082/tools

# For secure server, add API key header
curl -X POST -H "Content-Type: application/json" -H "X-API-Key: YOUR_API_KEY" -d '{"name": "list_sensors"}' http://localhost:8082/tools

Extending the Server

To add more tools:

1. Add the tool definition to `tools.json`

2. Implement the tool handler in `mcp_server.py` and `secure_mcp_server.py`

Production Considerations

For production deployment:

• Use HTTPS
• Deploy behind a reverse proxy like Nginx
• Implement more robust authentication (JWT, OAuth)
• Add rate limiting
• Set up persistent data storage for subscriptions

Deploying on Smithery.ai

The Fledge MCP Server can be deployed on Smithery.ai for enhanced scalability and availability. Follow these steps to deploy:

1. Prerequisites

2. Build and Deploy

# Build the Docker image
   docker build -t fledge-mcp .

   # Deploy to Smithery.ai
   smithery deploy

3. Configuration

The `smithery.json` file contains the configuration for your deployment:

4. Environment Variables

Set the following environment variables in your Smithery.ai dashboard:

5. Verification

After deployment, verify your server is running:

smithery status fledge-mcp

6. Monitoring

Monitor your deployment through the Smithery.ai dashboard:

7. Updating

To update your deployment:

# Build new image
   docker build -t fledge-mcp .
   
   # Deploy updates
   smithery deploy --update

JSON-RPC Protocol Support

The server implements the Model Context Protocol (MCP) using JSON-RPC 2.0 over WebSocket. The following methods are supported:

1. initialize

{
       "jsonrpc": "2.0",
       "method": "initialize",
       "params": {},
       "id": "1"
   }

Response:

{
       "jsonrpc": "2.0",
       "result": {
           "serverInfo": {
               "name": "fledge-mcp",
               "version": "1.0.0",
               "description": "Fledge Model Context Protocol (MCP) Server",
               "vendor": "Fledge",
               "capabilities": {
                   "tools": true,
                   "streaming": true,
                   "authentication": "api_key"
               }
           },
           "configSchema": {
               "type": "object",
               "properties": {
                   "fledge_api_url": {
                       "type": "string",
                       "description": "Fledge API URL",
                       "default": "http://localhost:8081/fledge"
                   }
               }
           }
       },
       "id": "1"
   }

2. tools/list

{
       "jsonrpc": "2.0",
       "method": "tools/list",
       "params": {},
       "id": "2"
   }

Response: Returns the list of available tools and their parameters.

3. tools/call

{
       "jsonrpc": "2.0",
       "method": "tools/call",
       "params": {
           "name": "get_sensor_data",
           "parameters": {
               "sensor_id": "temp1",
               "limit": 10
           }
       },
       "id": "3"
   }

Error Codes

The server follows standard JSON-RPC 2.0 error codes:

• -32700: Parse error
• -32600: Invalid Request
• -32601: Method not found
• -32602: Invalid params
• -32000: Server error