home / mcp / mcp platform mcp server

MCP Platform MCP Server

A scalable MCP-based platform exposing domain tools via a chat UI, with authoritative execution and audit capabilities.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "darkshloser-mcp": {
      "command": "python",
      "args": [
        "-m",
        "mcp_server.main"
      ],
      "env": {
        "LLM_MODEL": "gpt-4",
        "PYTHONPATH": "$PWD/src",
        "LLM_API_KEY": "YOUR_API_KEY",
        "LLM_API_BASE": "https://api.openai.com/v1",
        "LLM_PROVIDER": "azure_openai",
        "MCP_LOG_LEVEL": "INFO",
        "MCP_ENVIRONMENT": "development",
        "MCP_SERVER_PORT": "8001",
        "ORCHESTRATOR_PORT": "8000"
      }
    }
  }
}

You can operate a scalable MCP-based platform that interprets user intents via an LLM and executes domain-specific actions through an authoritative MCP server. This setup enables safe, isolated interactions with business domains such as HR, ERP, and DevOps through a chat interface.

How to use

You interact with the system using a chat UI. Your messages are managed by an orchestrator that talks to an LLM to interpret intent. The MCP server exposes domain-specific tools, enforces authorization, routes actions to the appropriate domain adapters, and audits every operation. As a user, you submit requests through natural language, the system translates them into tool calls, and the MCP server ensures actions align with your permissions and domain rules.

How to install

Follow these concrete steps to set up the MCP platform locally.

# Prerequisites
# - Python 3.10+
# - Node.js 18+
# - Docker (optional)

# 1. Clone the project and prepare environment
cd mcp_poc
cp .env.example .env
# Edit .env with your LLM API keys and any required settings

# 2. Install Python dependencies (from project root)
pip install -e ".[dev]"

# 3. Start MCP Server (open a new terminal)
export PYTHONPATH=$PWD/src
python -m mcp_server.main

# 4. Start Orchestrator (open another terminal)
export PYTHONPATH=$PWD/src
python -m orchestrator.main

# 5. Start Frontend (open another terminal)
cd frontend
npm install
npm run dev

# 6. Open browser
# - Frontend: http://localhost:3000
```

```
# Optional Docker-based setup
# Build and run all services
docker-compose up --build

# Access endpoints
# - Frontend: http://localhost:3000
# - Orchestrator API: http://localhost:8000
# - MCP Server API: http://localhost:8001

Configuration

Configure runtime behavior through environment variables and configuration files. The system uses environment-based settings for runtime control, and domain configurations are loaded via YAML settings.

Security and governance

Security is enforced at the MCP Server level. You authenticate clients, and the server authorizes requests using role-based and scope-based access control. All tool executions are audited, and sensitive parameters are automatically redacted. The orchestrator manages user identity and passes validated tool calls to the MCP server for authoritative execution.

Examples and domain capabilities

The platform includes three application domains with a set of tools you can invoke through the MCP server.

Testing and development

Use the test suite to validate behavior and coverage during development.

Notes

Domain isolation is fundamental: each domain operates in isolation with its own tool definitions, adapters, and permissions. New domains can be added through configuration alone, without changing core MCP server code.

Available tools

hr.get_employee

Fetches employee details by ID.

hr.search_employees

Searches employees by department and query.

hr.list_departments

Lists all departments.

erp.get_invoice

Retrieves an invoice by its ID.

erp.create_invoice

Creates a new invoice with customer and item details.

erp.check_low_stock

Checks for low stock in a given category.

devops.list_pods

Lists pods in a specified namespace.

devops.get_pod_logs

Retrieves logs for a specific pod.

devops.scale_deployment

Scales a deployment to a desired number of replicas.