otak-mcp-servers

This is a test project for a Model Context Protocol (MCP) server built using Cloudflare Workers and Durable Objects. It supports both Server-Sent Events (SSE) and Streamable HTTP transports.

Overview

This project provides a basic implementation example of an MCP server. It runs on Cloudflare Workers, utilizes Durable Objects to manage state, and provides MCP tools. SQLite storage is enabled.

Features

• MCP Server: Implementation of an MCP server using @modelcontextprotocol/sdk.
• Transports:
  • Server-Sent Events (SSE) (/sse endpoint)
  • Streamable HTTP (/mcp endpoint)
• Provided Tools:
  • dice_roll: Returns the result of rolling a die with the specified number of sides (default is 6).
    • Input Schema: { sides?: number } (1-100)
  • weather: Returns weather information (currently mock data) for the specified city.
    • Input Schema: { city: string }

Tech Stack

• Cloudflare Workers
• Cloudflare Durable Objects (SQLite backend)
• TypeScript
• Zod (for schema validation)
• @modelcontextprotocol/sdk
• agents (McpAgent)
• Wrangler (Development and deployment tool)

Setup and Execution

1. Install Dependencies

Navigate to the project directory (otak-mcp-test) and install the dependencies.

cd otak-mcp-test
npm install

2. Local Development

Start the local development server using Wrangler.

npm start
# or
npx wrangler dev

This runs the Worker locally and automatically reflects changes.

3. Deployment

Deploy the Worker to Cloudflare.

npm run deploy
# or
npx wrangler deploy

4. SSE Test

Use the provided script (test-sse.js) to test the SSE endpoint of the local or deployed Worker.

npm run test:sse <URL>

Replace <URL> with the URL of the local development server (e.g., http://localhost:8787/sse) or the deployed Worker URL (e.g., https://your-worker-name.your-subdomain.workers.dev/sse).

Endpoints

• SSE: https://<YOUR_WORKER_URL>/sse
• Streamable HTTP: https://<YOUR_WORKER_URL>/mcp

Replace <YOUR_WORKER_URL> with the URL of your deployed Worker.

Configuration

Worker configuration is done in the otak-mcp-test/wrangler.jsonc file. Key settings include:

• name: Worker name
• main: Entry point file (src/index.ts)
• compatibility_date, compatibility_flags: Worker compatibility settings
• durable_objects: Durable Object binding settings (MCP_OBJECT is bound to the MyMCP class)
• migrations: Durable Object migration settings (e.g., enabling SQLite)

Source Code (otak-mcp-test/src)

• index.ts:
  • The main entry point for the Cloudflare Worker.
  • Handles incoming requests and routes them based on the path (/sse or /mcp).
  • Delegates request processing to the MyMCP Durable Object.
• MyMcp.ts:
  • Implementation of the MyMCP Durable Object class.
  • Extends McpAgent and initializes McpServer.
  • Defines the available MCP tools (dice_roll, weather) and handles their execution logic.

Debugging

To view real-time logs from the deployed Worker, use the wrangler tail command.

npx wrangler tail otak-mcp-test --format pretty
# Or for detailed JSON format
npx wrangler tail otak-mcp-test --format json

MCP Inspector

The @modelcontextprotocol/inspector GUI tool is useful for testing the MCP server.

npx @modelcontextprotocol/inspector

After launching, access it in your browser at http://127.0.0.1:6274.

• Transport Type: Select SSE or Streamable HTTP.
• URL: Enter the Worker's endpoint URL (e.g., https://<YOUR_WORKER_URL>/sse or https://<YOUR_WORKER_URL>/mcp).
• Connect: Click to connect.
• List Tools: View the list of available tools.
• Run Tool: Execute a tool with specified parameters.

For more detailed debugging techniques, refer to .roo/rules/1-debug.md.

otak-mcp-pmbok

This project is an MCP server designed to answer questions about the Project Management Body of Knowledge (PMBOK) using a Retrieval-Augmented Generation (RAG) approach. It leverages Cloudflare Workers AI and Vectorize.

Overview

Based on the otak-mcp-test structure, this Worker imports PMBOK content (from otak-mcp-pmbok/pmbok.md), chunks and vectorizes it upon the first request, and stores the embeddings in a Cloudflare Vectorize index (pmbok-index). It provides an MCP tool (ask_pmbok) to query this knowledge base.

Features

• MCP Server: Implementation using @modelcontextprotocol/sdk.
• Transports: Supports Streamable HTTP (/mcp). SSE might work but is untested for this specific RAG implementation.
• RAG Pipeline:
  • Uses Cloudflare Workers AI (@cf/baai/bge-base-en-v1.5) for text embeddings.
  • Uses Cloudflare Vectorize for similarity search.
  • Uses Cloudflare Workers AI (@cf/meta/llama-3-8b-instruct) for answer generation based on retrieved context.
• Provided Tools:
  • ask_pmbok: Answers questions about PMBOK based on the content in pmbok.md.
    • Input Schema: { query: string }

Setup and Execution

Similar to otak-mcp-test, but navigate to the otak-mcp-pmbok directory.

1. Install Dependencies

cd otak-mcp-pmbok
npm install

2. Create Vectorize Index (if not done automatically by deploy)

The Worker expects a Vectorize index named pmbok-index. If the first deployment fails due to a missing index, create it manually:

# Ensure you are in the otak-mcp-pmbok directory
npx wrangler vectorize create pmbok-index --dimensions=768 --metric=cosine

(Note: The embedding model @cf/baai/bge-base-en-v1.5 outputs 768 dimensions, and cosine is suitable for text similarity.)

3. Deployment

# Ensure you are in the otak-mcp-pmbok directory
npm run deploy
# or
npx wrangler deploy

The first request after deployment will trigger the vectorization and indexing process, which might take some time.

4. Testing

Use the MCP Inspector (npx @modelcontextprotocol/inspector) to connect to the deployed Worker's /mcp endpoint (e.g., https://otak-mcp-pmbok.your-subdomain.workers.dev/mcp) and test the ask_pmbok tool.

Configuration

Configuration is managed in otak-mcp-pmbok/wrangler.toml. Key additions include:

• rules: Defines how to load .md files as text.
• ai: Binding for Cloudflare Workers AI.
• vectorize: Binding for the pmbok-index Vectorize index.

Source Code (otak-mcp-pmbok/src)

• index.ts: Main Worker entry point, routes requests to the Durable Object.
• MyMcp.ts: Implements the Durable Object, MCP server, RAG pipeline, and the ask_pmbok tool. Handles dynamic vectorization and indexing on first load.
• types/assets.d.ts: TypeScript definitions for importing .md files.

Debugging

Use wrangler tail to view real-time logs:

npx wrangler tail otak-mcp-pmbok --format pretty

otak-mcp-commander

This project is a C# implementation of an MCP server that provides various system command and file operation tools using the ModelContextProtocol library. It uses stdio as its transport mechanism, making it suitable for embedded use cases where direct process communication is needed.

Overview

otak-mcp-commander is a .NET-based MCP server that runs as a console application and communicates via standard input/output. It provides several utility tools for file operations, command execution, and logging that can be invoked through the Model Context Protocol.

Features

• MCP Server: Implementation using the ModelContextProtocol .NET library.
• Transport: Stdio-based communication (WithStdioServerTransport).
• Provided Tools:
  • GetCurrentDirectory: Returns the current working directory.
  • ListFiles: Lists files and directories in a specified path.
  • ExecuteCommand: Executes a command-line command and returns its output.
  • ReadFiles: Reads the content of a specified file.

Tech Stack

• .NET 9.0 (Runtime and SDK)
• ASP.NET Core (Web framework)
• ModelContextProtocol (MCP library for .NET)

Setup and Execution

1. Prerequisites

• .NET 9.0 SDK or later installed

2. Build the Project

Navigate to the project directory and build the application:

cd otak-mcp-commander
dotnet build

3. Run the MCP Server

dotnet run

This starts the MCP server, which listens on stdin for incoming requests and responds via stdout.

4. Connect to the Server

To communicate with the server, you need to write a client that can send JSON-RPC formatted messages to the server's stdin and read responses from its stdout. The server expects JSON-RPC 2.0 format messages.

Example request format:

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

Source Code

• Program.cs: Main entry point that configures and starts the MCP server.
• CommanderTool.cs: Implements the MCP tools and their execution logic.