HPE GreenLake MCP Servers

A collection of Model Context Protocol (MCP) servers that enable AI assistants to interact with HPE GreenLake platform services.

Table of Contents

• What is the Model Context Protocol?
• What's in this Repository?
• Available MCP Servers
• Quick Start
• MCP Client Configuration
• Tool Modes
• Architecture Overview
• Security Considerations
• Troubleshooting
• Getting Help
• Resources
• License

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open protocol that enables seamless integration between LLM applications and external data sources and tools. Whether you're building an AI-powered IDE, enhancing a chat interface, or creating custom AI workflows, MCP provides a standardized way to connect LLMs with the context they need.

What's in this Repository?

This repository contains production-ready MCP servers for HPE GreenLake platform services. Each server provides AI assistants with secure, read-only access to specific GreenLake APIs, enabling natural language interactions with your HPE infrastructure.

Key Features

• Standardized Architecture: All servers follow consistent patterns for authentication, configuration, and error handling
• OAuth2 Authentication: Secure access using HPE GreenLake workspace credentials with automatic token management
• Dual Tool Modes: Each server supports both static (individual tools per endpoint) and dynamic (meta-tools) operation modes
• Type-Safe: Built with Pydantic models for runtime validation and type safety
• Production Ready: Comprehensive logging, error handling, and Docker support
• Well Tested: Extensive unit and integration test coverage

Available MCP Servers

Quick Start

Prerequisites

• Python 3.10 or higher
• uv package manager
• HPE GreenLake workspace with API credentials

Configuration

All servers require the following HPE GreenLake credentials:

export GREENLAKE_API_BASE_URL="https://global.api.greenlake.hpe.com"
export GREENLAKE_CLIENT_ID="your-client-id"
export GREENLAKE_CLIENT_SECRET="your-client-secret"
export GREENLAKE_WORKSPACE_ID="your-workspace-id"

Running a Server

# Navigate to a server directory
cd src/audit-logs

# Install dependencies
uv sync

# Run the server
uv run python -m greenlake_audit_logs_mcp

MCP Client Configuration

Claude Desktop

Add servers to your claude_desktop_config.json:

{
  "mcpServers": {
    "greenlake-audit-logs": {
      "command": "uv",
      "args": ["run", "python", "-m", "greenlake_audit_logs_mcp"],
      "cwd": "/path/to/gl-mcp/src/audit-logs",
      "env": {
        "GREENLAKE_API_BASE_URL": "https://global.api.greenlake.hpe.com",
        "GREENLAKE_CLIENT_ID": "your-client-id",
        "GREENLAKE_CLIENT_SECRET": "your-client-secret",
        "GREENLAKE_WORKSPACE_ID": "your-workspace-id"
      }
    },
    "greenlake-devices": {
      "command": "uv",
      "args": ["run", "python", "-m", "greenlake_devices_mcp"],
      "cwd": "/path/to/gl-mcp/src/devices",
      "env": {
        "GREENLAKE_API_BASE_URL": "https://global.api.greenlake.hpe.com",
        "GREENLAKE_CLIENT_ID": "your-client-id",
        "GREENLAKE_CLIENT_SECRET": "your-client-secret",
        "GREENLAKE_WORKSPACE_ID": "your-workspace-id"
      }
    }
  }
}

Other MCP Clients

These servers use the standard stdio transport and work with any MCP-compatible client. See individual server READMEs for specific configuration examples.

Tool Modes

All servers support two operation modes that can be switched at runtime using the MCP_TOOL_MODE environment variable:

Static Mode (Default)

• Individual tools: Each API endpoint becomes a dedicated MCP tool
• Type-safe: Explicit tool definitions with compile-time validation
• Discoverable: Tools appear individually in MCP client interfaces
• Best for: Smaller APIs and focused workflows

export MCP_TOOL_MODE=static

Dynamic Mode

• Meta-tools: 3 generic tools that can handle any API endpoint
  • list_endpoints - Discover available API endpoints
  • get_endpoint_schema - Get detailed schema information
  • invoke_dynamic_tool - Execute API calls with runtime validation
• Memory efficient: Lower overhead for large APIs
• Best for: Large APIs with many endpoints

export MCP_TOOL_MODE=dynamic

Compatibility Matrix

Architecture Overview

High-Level Architecture

flowchart TB
    subgraph clients["<b> MCP Clients </b>"]
        direction TB
        claude["Claude Desktop"]
        vscode["VS Code"]
        other["..."]
    end

    subgraph servers["<b> MCP Servers </b>"]
        direction TB
        audit["Audit Logs"]
        devices["Devices"]
        users["Users"]
        workspaces["Workspaces"]
        subscriptions["Subscriptions"]
        reporting["Reporting"]
        serviceX["..."]
        
        subgraph shared["<i> Shared Components </i>"]
            direction LR
            oauth["OAuth2 Token Manager<br/>(with caching)"]
        end
    end

    subgraph platform["<b>&nbsp; HPE GreenLake Platform &nbsp;</b>"]
        direction TB
        api["APIs"]
    end

    clients ==>|JSON-RPC/stdio| servers
    servers ==>|HTTPS/REST + OAuth2| platform

    classDef clientStyle fill:#2196F3,stroke:none,color:#fff,rx:10,ry:10
    classDef serverStyle fill:#4CAF50,stroke:none,color:#fff,rx:10,ry:10
    classDef authStyle fill:#FF9800,stroke:none,color:#fff,rx:10,ry:10
    classDef platformStyle fill:#01a982,stroke:none,color:#fff,rx:10,ry:10
    classDef clientContainerStyle fill:none,stroke:#2196F3,stroke-width:3px,rx:15,ry:15
    classDef serverContainerStyle fill:none,stroke:#4CAF50,stroke-width:3px,rx:15,ry:15
    classDef authContainerStyle fill:none,stroke:#FF9800,stroke-width:2px,stroke-dasharray:5 5,rx:10,ry:10
    classDef platformContainerStyle fill:none,stroke:#01a982,stroke-width:3px,rx:15,ry:15
    
    class claude,vscode,other clientStyle
    class audit,devices,users,workspaces,subscriptions,reporting,serviceX serverStyle
    class oauth authStyle
    class api platformStyle
    class clients clientContainerStyle
    class servers serverContainerStyle
    class shared authContainerStyle
    class platform platformContainerStyle

Request Flow

1. Client Request: MCP client sends tool invocation via JSON-RPC over stdio
2. Authentication: Server authenticates with HPE GreenLake using OAuth2 (tokens cached and auto-refreshed)
3. API Call: Server makes authenticated HTTPS request to GreenLake API
4. Response: Server processes response and returns data to client via MCP protocol

Shared Components

All servers share a common architecture:

src/<service>/
├── greenlake_<service>_mcp/   # Main package (e.g. greenlake_audit_logs_mcp)
│   ├── __main__.py            # Entry point (python -m greenlake_<service>_mcp)
│   ├── auth/                  # OAuth2 authentication
│   ├── config/                # Settings and logging
│   ├── models/                # Pydantic data models
│   ├── server/                # MCP server core
│   ├── tools/                 # Tool implementations
│   │   ├── base.py           # Base tool class
│   │   ├── registry.py       # Tool registration
│   │   └── implementations/  # Tool implementations
│   └── utils/                 # HTTP client and utilities
├── tests/                     # Unit and integration tests
└── pyproject.toml             # Package definition and entry points

Authentication

All servers use OAuth2 client credentials flow with automatic token management:

• Tokens are cached and automatically refreshed
• Thread-safe token management
• Configurable retry logic with exponential backoff

Logging

Servers use structured logging with MCP protocol compliance:

• stderr: Diagnostic logs (controlled by GREENLAKE_LOG_LEVEL)
• File logging: Optional detailed logging to ~/.hpe/mcp-logs/
• stdout: Reserved for JSON-RPC messages only

Security Considerations

• Read-Only Access: All servers provide read-only API access by design
• Credential Management: Never commit credentials to version control
• Token Security: OAuth2 tokens are cached in memory only
• Workspace Isolation: Operations are scoped to the configured workspace
• HTTPS Only: All API communications use TLS encryption

Troubleshooting

Common Issues

Server won't start:

• Verify all required environment variables are set
• Check that uv is installed and dependencies are synced
• Review stderr output for configuration errors

Authentication failures:

• Verify client credentials are valid for your workspace
• Ensure GREENLAKE_WORKSPACE_ID matches your credential scope
• Check network connectivity to GreenLake APIs

MCP client connection issues:

• Verify the cwd path in client configuration is correct
• Check that the server starts successfully when run manually
• Review client logs for specific error messages

Debug Logging

Enable detailed logging for troubleshooting:

export GREENLAKE_LOG_LEVEL=DEBUG
export GREENLAKE_FILE_LOGGING=true

Log files are written to: ~/.hpe/mcp-logs/{service-name}/

Getting Help

GitHub Issues

The primary support channel for this project is GitHub Issues. When reporting an issue, please include:

• Environment details: OS, Python version, uv version
• Server name: Which MCP server you're using (audit-logs, devices, etc.)
• Error messages: Complete error output with stack traces
• Configuration: Relevant environment variable settings (DO NOT include credentials)
• Steps to reproduce: Detailed description of how to reproduce the issue

Before Opening an Issue

1. Check existing issues: Search open and closed issues
2. Review troubleshooting: Check the Troubleshooting section above
3. Enable debug logging: Run with GREENLAKE_LOG_LEVEL=DEBUG and GREENLAKE_FILE_LOGGING=true to gather detailed logs
4. Test authentication: Verify your credentials work with the GreenLake APIs directly

Issue Types

• Bug Report: For unexpected behavior, errors, or crashes
• Feature Request: For suggesting new capabilities or improvements
• Question: For clarification on usage, configuration, or architecture
• Documentation: For reporting documentation issues or suggesting improvements

Resources

• Model Context Protocol Documentation
• HPE GreenLake API Documentation
• MCP Specification

License

This project is licensed under the Apache 2.0 - see the LICENSE file for details.