DVMCP workshop

This workshop will guide you through exploring the concepts behind MCP servers and how to deploy them as DVMs in Nostr using DVMCP. By the end, you'll understand how these systems work together and be able to create your own deployments.

Understanding MCP Systems

MCP (Model Context Protocol) systems consist of two main components that work together:

1. MCP Server: The heart of the system that exposes tools, which you can access via the .listTools() method.
2. MCP Client: The interface that connects to the MCP server and lets you use the tools it offers.

These servers and clients can communicate using different transport methods:

• Standard I/O (stdio): A simple local connection method when your server and client are on the same machine.
• Server-Sent Events (SSE): Uses HTTP to create a communication channel.

For this workshop, we'll use stdio to deploy our server. DVMCP will act as a bridge, connecting to your MCP server as an MCP client, and exposing its tools as a DVM that anyone can call from Nostr.

Creating (or Finding) an MCP Server

Building an MCP server is simpler than you might think:

1. Create software in any programming language you're comfortable with.
2. Add an MCP library to expose your server's MCP interface.
3. Create an API that wraps around your software's functionality.

Once your server is ready, an MCP client can connect, for example, with bun index.js, and then call .listTools() to discover what your server can do. This pattern, known as reflection, makes Nostr DVMs and MCP a perfect match since both use JSON, and DVMs can announce and call tools, effectively becoming an MCP proxy.

Alternatively, you can use one of the many existing MCP servers available in various repositories.

For more information about mcp and how to build mcp servers you can visit https://modelcontextprotocol.io/

Setting Up the Workshop

Let's get hands-on:

First, to follow this workshop you will need Bun. Install it from https://bun.sh/. For Linux and macOS, you can use the installation script:

curl -fsSL https://bun.sh/install | bash

1. Choose your MCP server: You can either create one or use an existing one.

2. Inspect your server using the MCP inspector tool:

   npx @modelcontextprotocol/inspector build/index.js arg1 arg2
   

   This will:

   • Launch a client UI (default: http://localhost:5173)
   • Start an MCP proxy server (default: port 3000)
   • Pass any additional arguments directly to your server

3. Use the inspector: Open the client UI in your browser to connect with your server, list available tools, and test its functionality.

Deploying with DVMCP

Now for the exciting part – making your MCP server available to everyone on Nostr:

1. Navigate to your MCP server directory.

2. Run without installing (quickest way):

   npx @dvmcp/bridge
   

3. Or install globally for regular use:

   npm install -g @dvmcp/bridge
   # or
   bun install -g @dvmcp/bridge
   

   Then run using:

   dvmcp-bridge
   

This will guide you through creating the necessary configuration.

Watch the console logs to confirm successful setup – you'll see your public key and process information, or any issues that need addressing.

For the configuration, you can set the relay as wss://relay.dvmcp.fun , or use any other of your preference

Testing and Integration

1. Visit dvmcp.fun to see your DVM announcement.
2. Call your tools and watch the responses come back.

For production use, consider running dvmcp-bridge as a system service or creating a container for greater reliability and uptime.

Integrating with LLM Clients

You can also integrate your DVMCP deployment with LLM clients using the discovery package:

1. Install and use the @dvmcp/discovery package:

   npx @dvmcp/discovery
   

2. This package acts as an MCP server for your LLM system by:

   • Connecting to configured Nostr relays
   • Discovering tools from DVMCP servers
   • Making them available to your LLM applications

3. Connect to specific servers or providers using these flags:

   # Connect to all DVMCP servers from a provider
   npx @dvmcp/discovery --provider npub1...
   
   # Connect to a specific DVMCP server
   npx @dvmcp/discovery --server naddr1...
   

Using these flags, you wouldn't need a configuration file. You can find these commands and Claude desktop configuration already prepared for copy and paste at dvmcp.fun.

This feature lets you connect to any DVMCP server using Nostr and integrate it into your client, either as a DVM or in LLM-powered applications.

Final thoughts

If you've followed this workshop, you now have an MCP server deployed as a Nostr DVM. This means that local resources from the system where the MCP server is running can be accessed through Nostr in a decentralized manner. This capability is powerful and opens up numerous possibilities and opportunities for fun.

You can use this setup for various use cases, including in a controlled/local environment. For instance, you can deploy a relay in your local network that's only accessible within it, exposing all your local MCP servers to anyone connected to the network. This setup can act as a hub for communication between different systems, which could be particularly interesting for applications in home automation or other fields. The potential applications are limitless.

However, it's important to keep in mind that there are security concerns when exposing local resources publicly. You should be mindful of these risks and prioritize security when creating and deploying your MCP servers on Nostr.

Finally, these are new ideas, and the software is still under development. If you have any feedback, please refer to the GitHub repository to report issues or collaborate. DVMCP also has a Signal group you can join. Additionally, you can engage with the community on Nostr using the #dvmcp hashtag.

Useful Resources

• Official Documentation:

  • Model Context Protocol: modelcontextprotocol.org
  • DVMCP.fun: dvmcp.fun

• Source Code and Development:

  • DVMCP: github.com/gzuuus/dvmcp
  • DVMCP.fun: github.com/gzuuus/dvmcpfun

• MCP Servers and Clients:

  • Smithery AI: smithery.ai
  • MCP.so: mcp.so
  • Glama AI MCP Servers: glama.ai/mcp/servers

• Signal group

Happy building!