Dpdispatcher MCP Server

🚀 Dispatcher MCP Server

A Model Context Protocol (MCP) server wrapped by the dpdispatcher library. It enables language models or other MCP clients to submit and manage computational tasks on local machines or HPC clusters supported by dpdispatcher.

🚀 Quick Start

The Dispatcher MCP Server allows you to submit and manage computational tasks through the MCP protocol. Here's how to get started:

Prerequisites

• Python 3.x
• Installed dpdispatcher library (pip install dpdispatcher)
• Installed mcp library (pip install mcp)
• Installed anyio library (pip install anyio)

Clone/Place Files

Ensure that fast_server.py and job_manager.py (along with __init__.py) are in a directory (e.g., dispatcher_mcp_server).

Configure dpdispatcher

If you want to submit tasks to a remote HPC or Bohrium, make sure dpdispatcher itself is correctly configured (e.g., SSH keys, Bohrium credentials).

Run the Server

Navigate to the parent directory containing dispatcher_mcp_server and run:

python dispatcher_mcp_server/fast_server.py

The server will start and listen through standard input and output.

Integrate with MCP Clients (e.g., Cline)

Add the following configuration to the client's MCP settings (e.g., mcp_settings.json for Cline), adjusting the paths as needed:

{
  "mcpServers": {
    "dispatcher-mcp-server": {
      "command": "python",
      "args": [
        "dispatcher_mcp_server/fast_server.py"
      ],
      "cwd": "/path/to/parent/directory/containing/dispatcher_mcp_server",
      "env": {
        "PYTHONPATH": "/path/to/parent/directory/containing/dispatcher_mcp_server"
      },
      "disabled": false
    }
  }
}

Restart the client to load the server.

✨ Features

• Exposes dpdispatcher functionality through standard MCP tools.
• submit_job: Submit new computational tasks.
• query_status: Check the status of submitted tasks.
• cancel_job: Attempt to cancel running or queued tasks.
• fetch_result: Retrieve the result file paths of completed tasks.
• Includes MCP resources and prompts to guide interactive task configuration.
• Supports standard input-output transfer (stdio) for local integration (e.g., with Cline).

📦 Installation

1. Prerequisites:
   • Python 3.x
   • Install dpdispatcher library: pip install dpdispatcher
   • Install mcp library: pip install mcp
   • Install anyio library: pip install anyio
2. Clone/Place Files: Make sure fast_server.py and job_manager.py (along with __init__.py) are in a directory (e.g., dispatcher_mcp_server).
3. Configure dpdispatcher: If submitting to a remote HPC or Bohrium, ensure dpdispatcher is properly configured (e.g., SSH keys, Bohrium credentials).

💻 Usage Examples

Basic Usage

Use an MCP client to interact with the server. You can directly call the submit_job tool, providing the necessary parameters (machine configuration, resource configuration, task details), or use the configure_job prompt to guide the LLM through an interactive configuration process. There are also auxiliary resources available, such as dpd://examples/machine/{type} for context reference.

# Example of running the server
python dispatcher_mcp_server/fast_server.py

// Example of MCP client configuration
{
  "mcpServers": {
    "dispatcher-mcp-server": {
      "command": "python",
      "args": [
        "dispatcher_mcp_server/fast_server.py"
      ],
      "cwd": "/path/to/parent/directory/containing/dispatcher_mcp_server",
      "env": {
        "PYTHONPATH": "/path/to/parent/directory/containing/dispatcher_mcp_server"
      },
      "disabled": false
    }
  }
}