MCP4DRL

Exposes a trained DQN agent for business process allocation via conversational queries and explainability features.

other

0

GitHub Stars

other

Language

4 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "mostapow-mcp4drl": {
      "command": "cmd.exe",
      "args": [
        "/c",
        "C:\\path\\to\\mcp4drl_repo\\run_server.bat"
      ]
    }
  }
}

You can run the MCP4DRL server to expose a trained reinforcement learning agent for real-time, language-driven resource allocation in business processes. This server lets you query the environment state, inspect Q-values, receive action recommendations with justifications, compare against simple heuristics, and control simulations through natural language interfaces.

How to use

To use the MCP4DRL server, run the local MCP server executable on your machine and connect your MCP client. Your client can query the current simulation state, view eligible actions, fetch Q-values for all actions, ask for the agent’s top action with justification, and request explanations for decisions. You can also compare the RL policy against common baselines and control the simulation by stepping through steps, resetting, or running full episodes.

How to install

Prerequisites you need before installing and running the server:

Install required Python packages from the provided requirements file.

pip install -r requirements.txt

Additional setup and quick start

Test the integration locally to ensure the server and client can communicate, then start the MCP server.

python -m mcp4drl.test_integration

Start the MCP server on your machine

Run the server using the platform-specific launcher shown here. Use the Windows launcher if you are on Windows, or the Unix launcher if you are on Linux or macOS.

# Windows
run_server.bat

# Linux/Mac
chmod +x run_server.sh
./run_server.sh

Available tools

get_environment_state

Retrieve the current simulation state, including environment variables, waiting and active cases, and available resources.

get_eligible_actions

Fetch all possible actions and indicate their validity given the current state.

get_q_values

Obtain Q-values for every possible action to understand the agent's preferences.

get_recommended_action

Request the agent’s top action along with justification.

explain_action

Receive a detailed explanation for why a particular action was chosen.

compare_with_heuristic

Compare the agent’s decisions against common heuristics such as FIFO, SPT, EDF, and LST.

step_simulation

Advance the simulation by a single step to observe immediate outcomes.

reset_simulation

Reset the environment to its initial state for fresh experiments.

run_episode

Run a full episode using the current policy to evaluate long-run performance.