Docs & Usage Guide

OpenStack MCP Server

Overview

An OpenStack resource query service based on MCP (Model Context Protocol), providing API interfaces to query compute, storage, network, image, and other resources from the OpenStack cloud platform.

Features

Real-time Resource Queries: Get the latest resource status of OpenStack clusters via API
Multi-dimensional Information: Support querying various resources including compute, storage, network, and images
Flexible Filtering: Filter resources by name, ID, and other conditions
Detail Level Control: Support basic, detailed, and full information display levels
Standard MCP Interface: Fully compatible with MCP protocol, seamlessly integrating with large language models

Technical Architecture

High-performance asynchronous HTTP service based on Starlette and Uvicorn
Interact with OpenStack API using OpenStack SDK
Provide structured OpenStack resource information to large language models through MCP protocol
Support SSE streaming output for real-time feedback

Architecture Diagram

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│             │    │             │    │             │
│  AI Client  │───▶│  MCP Server │───▶│  OpenStack  │
│  (LLM)      │◀───│  (Server)   │◀───│  API        │
│             │    │             │    │             │
└─────────────┘    └─────────────┘    └─────────────┘

Installation

Requirements

Python 3.10+
OpenStack environment

Install via pip

pip install openstack-mcp-server

Usage

Start the Service

openstack-mcp-server --port 8000 --log-level INFO --auth-url 'http://<OpenStack-API-Endpoint>:5000/v3' --username '<OpenStack-Admin-User>' --password '<OpenStack-Admin-Password>'

After starting, the MCP interface will be available at http://localhost:8000/openstack.

Parameters

--port: Service listening port, default is 8000
--log-level: Log level, options are DEBUG, INFO, WARNING, ERROR, CRITICAL, default is INFO
--json-response: Use JSON response instead of SSE stream, default is False

Interface Example

Through the MCP protocol, you can use the following tools to query OpenStack resources:

Get OpenStack VM Instances

{
  "name": "get_instances",
  "arguments": {
    "filter": "web-server",
    "limit": 10,
    "detail_level": "detailed"
  }
}

Parameter description:

filter: Filter condition, such as instance name or ID (optional)
limit: Maximum number of results to return (optional, default 100)
detail_level: Level of detail in the returned information, options are basic, detailed, full (optional, default detailed)

Install from Source

# Clone repository
git clone https://github.com/wangshqly0407/openstack-mcp-server.git
cd openstack-mcp-server
# Create virtual environment
uv venv
# Activate virtual environment
source .venv/bin/activate
# Initialize runtime environment
uv sync
# Start streaming HTTP MCP server
uv run ./src/mcp_openstack_http/server.py --port 8000 --log-level INFO --auth-url 'http://<OpenStack-API-Endpoint>:5000/v3' --username '<OpenStack-Admin-User>' --password '<OpenStack-Admin-Password>'

Testing and Verification

# Method 1: Test using npx
npx -y @modelcontextprotocol/inspector uv run ./src/mcp_openstack_http/server.py --port 8000 --log-level INFO --auth-url 'http://<OpenStack-API-Endpoint>:5000/v3' --username '<OpenStack-Admin-User>' --password '<OpenStack-Admin-Password>'

# Method 2: Test using docker
docker run -it --rm -p 6274:6274 -p 6277:6277 -v $(pwd):/app -w /app node:18 npx -y @modelcontextprotocol/inspector uv run ./src/mcp_openstack_http/server.py --port 8000 --log-level INFO --auth-url 'http://<OpenStack-API-Endpoint>:5000/v3' --username '<OpenStack-Admin-User>' --password '<OpenStack-Admin-Password>'

Access: http://localhost:6274/

Development Extensions

Adding New Resource Query Tools

Add corresponding resource acquisition functions in src/mcp_openstack_http/server.py
Register the new tool in the list_tools method
Implement the tool's processing logic in the call_tool method

License

Apache 2.0

OpenStack