central-mcp-server

Community MCP server for HPE Aruba Networking Central. This exposes your Central data as tools that AI assistants can query directly.

Table of Contents

• Overview

• Getting Started

  • Getting Your Credentials

  • Installation

  • MCP Client Configuration

    • Claude Desktop

    • Claude Code

    • GitHub Copilot (VS Code)

    • HTTP Transport (Streamable HTTP)

• Supported Capabilities

• What You Can Ask

• Example Queries

• Dev Setup

Related MCP server: API-Central

Overview

central-mcp-server wraps Central REST APIs and exposes them as MCP (Model Context Protocol) tools — 12 MCP tools across inventory, monitoring, events, alerts, and troubleshooting. Once configured, AI assistants like Claude or GitHub Copilot can answer questions like:

• "Which sites have poor health scores right now?"

• "Show me all failed wireless clients at HQ in the last 24 hours."

• "Show me all online access points at the Chicago office."

• "What events happened on switch SW-CORE-01 yesterday?"

See the full overview guide for a deeper look at capabilities, limitations, and how the server works.

Getting Started

Getting Your Credentials

You need three values to connect this server to Central's REST APIs: CENTRAL_BASE_URL, CENTRAL_CLIENT_ID, and CENTRAL_CLIENT_SECRET.

API Gateway Base URL (CENTRAL_BASE_URL)

The API gateway base URL for your Central account (e.g. https://us5.api.central.arubanetworks.com).

For instructions on how to locate your base URL, see Finding Your Base URL in Central.

API Client Credentials (CENTRAL_CLIENT_ID & CENTRAL_CLIENT_SECRET)

OAuth credentials created through the HPE GreenLake Platform:

1. Log in to your HPE GreenLake account and open Manage Workspace.

2. Click Personal API clients.

3. Click Create Personal API client.

4. Give it a nickname (e.g. central-mcp-server) and select your HPE Aruba Networking Central instance from the service dropdown.

5. Click Create personal API client.

6. Copy both the Client ID and Client Secret immediately. The platform does not store the secret and it cannot be retrieved later.

Full guide: Generating and Managing Access Tokens

Installation

Install uv if you haven't already. It's the only prerequisite.

Using an MCP client (Claude Desktop, Claude Code, GitHub Copilot)? No install command needed. Jump to MCP Client Configuration, the client fetches and runs the server automatically via uvx.

Want the server as a persistent CLI tool on your PATH?

uv tool install --prerelease=allow central-mcp-server

--prerelease=allow is required because this server depends on pycentral, which currently only has a pre-release version on PyPI. uv skips pre-releases by default.

See the full setup guide for prerequisites, troubleshooting, and step-by-step instructions.

MCP Client Configuration

Replace the placeholder values with your actual credentials in all examples below.

Optional: Code Mode Transform (DYNAMIC_TOOLS)

DYNAMIC_TOOLS is optional and only affects startup behavior:

• Code Mode is enabled only when DYNAMIC_TOOLS is set to true (case-insensitive).

• Code Mode is disabled when DYNAMIC_TOOLS is not set or set to any other value.

• Variable name is strict: use DYNAMIC_TOOLS (plural). DYNAMIC_TOOL is ignored.

When enabled, the server starts with CodeMode() and exposes Code Mode meta-tools to the client. When disabled, the server runs without the transform and exposes the normal registered tool catalog directly. Recommended to use CodeMode() when you have multiple MCP servers running to preserve your context window.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "central-mcp": {
      "command": "uvx",
      "args": ["--prerelease=allow", "central-mcp-server"],
      "env": {
        "CENTRAL_BASE_URL": "your-central-base-url",
        "CENTRAL_CLIENT_ID": "your-client-id",
        "CENTRAL_CLIENT_SECRET": "your-client-secret"
      }
    }
  }
}

See the Claude Desktop setup guide for full steps and troubleshooting.

Claude Code

claude mcp add central-mcp \
  -e CENTRAL_BASE_URL=your-central-base-url \
  -e CENTRAL_CLIENT_ID=your-client-id \
  -e CENTRAL_CLIENT_SECRET=your-client-secret \
  -- uvx --prerelease=allow central-mcp-server

See the Claude Code setup guide for full steps and troubleshooting.

GitHub Copilot (VS Code)

Add .vscode/mcp.json to your workspace root and add that path to .gitignore to keep credentials out of version control:

{
  "servers": {
    "central-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--prerelease=allow", "central-mcp-server"],
      "env": {
        "CENTRAL_BASE_URL": "your-central-base-url",
        "CENTRAL_CLIENT_ID": "your-client-id",
        "CENTRAL_CLIENT_SECRET": "your-client-secret"
      }
    }
  }
}

Add to .gitignore:

.vscode/mcp.json

See the GitHub CoPilot setup guide for full steps and troubleshooting.

HTTP Transport (Streamable HTTP)

By default the server runs over stdio, which is the right choice for most MCP clients. If you need to run the server as a persistent HTTP process — for example, to share it across multiple clients or to connect via a remote URL — you can switch to the streamable-http transport.

Step 1 — Install the server as a CLI tool (if you haven't already):

uv tool install --prerelease=allow central-mcp-server

--prerelease=allow is required because this server depends on pycentral, which currently only has a pre-release version on PyPI.

Step 2 — Create a .env file in your working directory with your credentials and transport settings:

CENTRAL_BASE_URL=your-central-base-url
CENTRAL_CLIENT_ID=your-client-id
CENTRAL_CLIENT_SECRET=your-client-secret
MCP_TRANSPORT=http
MCP_HOST=127.0.0.1
MCP_PORT=8000

Step 3 — Start the server:

# If installed via uv tool install:
central-mcp-server

# If running from source:
python server.py

The MCP endpoint will be available at http://<MCP_HOST>:<MCP_PORT>/mcp.

Step 4 — Connect your MCP client to the running server:

claude mcp add central-mcp --transport http --url http://127.0.0.1:8000/mcp

Or add it to your MCP client config:

{
  "mcpServers": {
    "central-mcp": {
      "type": "http",
      "url": "http://127.0.0.1:8000/mcp"
    }
  }
}

Note: Credentials must be set in the server's environment (via .env or OS env vars) before starting it. They are not passed through the HTTP client config.

Supported Capabilities

The server covers eleven capability categories across the new Central REST API families (Network Monitoring, Network Notifications, Network Troubleshooting):

See the Capability Reference for every tool in each category, the Central API family it uses, and scope/limitations.

What You Can Ask

Once connected, you can ask your AI assistant questions like:

• "Give me a health overview of all sites."

• "Which sites are in poor health right now?"

• "Show me all access points at the Chicago office."

• "List the switches in the London campus and show CPU and PoE trends for SG34L5002Y."

• "How healthy is the BLR gateway cluster, and what's its client capacity trend?"

• "What critical alerts are active across the network?"

• "Find all failed wireless clients at HQ in the last 24 hours."

• "How is client onboarding doing today, and why are clients failing authentication?"

• "What events happened on switch SW-CORE-01 yesterday?"

• "Ping 8.8.8.8 from switch SW-CORE-01."

• "Run 'show version' and 'show interfaces brief' on switch SG43KN5017."

• "Bounce PoE on port 1/1/6 of switch SG43KN5017."

See Central MCP Server in Action for real query examples across all supported clients.

Example Queries

New to driving an AI assistant over your network? See What You Can Ask — a guided tour of real questions across every tool category (site health, devices, APs, switches, gateways, WLANs, clients, alerts, events, and live diagnostics), each shown with the answer it returns and a list of related questions to try.

Tool Categories

%%{init: {'theme':'base', 'themeVariables': {'background': '#0b0f1a'}}}%%
graph TD
    MCP["Central MCP Server"] --> Sites["Sites"]
    MCP --> Devices["Device Inventory"]
    MCP --> Clients["Clients"]
    MCP --> Alerts["Alerts"]
    MCP --> Events["Events"]
    MCP --> DevMon["Device Monitoring"]
    MCP --> WLAN["WLAN"]
    MCP --> Troubleshooting["Troubleshooting"]

    classDef mcp fill:#05cc93,color:#001b14,stroke:#000000,stroke-width:2px;
    classDef tool fill:#0070f8,color:#ffffff,stroke:#000000,stroke-width:1.5px;

    class MCP mcp;
    class Sites,Devices,DevMon,Clients,Alerts,Events,WLAN,Troubleshooting tool;

    linkStyle default stroke:#ffffff,stroke-width:2px;

Tools

The 0.2.x surface folds related operations into 13 tools. Envelope-returning reads default to response_format="concise"; use "detailed" for full item fields.

LLM Workflow for Events

Use this sequence for faster, lower-token event investigations:

1. For site-level queries, call events tools with site_id only.

2. For device/client queries, pass site_id plus context_type and context_identifier.

3. Call central_get_events with mode="facets" and response_mode="compact" to get ranked event names, source types, and categories.

4. Pick the top category/source/event name as your likely starting point.

5. Call central_get_events with mode="records" and targeted filters to fetch detailed records.

6. Use mode="facets", response_mode="full" only when exact per-value counts are required.

Guided Prompts

The server includes 12 built-in prompts to help AI assistants run common workflows:

Dev Setup

git clone <Github Server URL>
cd central-mcp-server

Create and activate a virtual environment, then install dependencies:

python3 -m venv .venv
source .venv/bin/activate
uv sync

Create .env with your credentials:

CENTRAL_BASE_URL=your-central-base-url
CENTRAL_CLIENT_ID=your-client-id
CENTRAL_CLIENT_SECRET=your-client-secret

Run the server:

python3 server.py

To install and test the package locally before publishing:

uv tool install .