Integrations

Three first-class ways to talk to the server: an AI desktop client (Claude), an interactive testing harness (MCP Inspector), and a REST shim (MCPO) for anything that doesn't speak MCP natively.

Claude Desktop

Claude Desktop reads its MCP server list from claude_desktop_config.json — ~/Library/Application Support/Claude/ on macOS, and %APPDATA%\Claude\ on Windows. Add the server, restart, and Claude will discover the tools on its next session.

{
  "mcpServers": {
    "mikrotik": {
      "command": "uvx",
      "args": [
        "mcp-server-mikrotik",
        "--host",     "<HOST>",
        "--username", "<USERNAME>",
        "--password", "<PASSWORD>",
        "--port",     "22"
      ]
    }
  }
}

Verifying the connection

After restarting Claude Desktop, look for the connector indicator in the prompt box. A first easy test:

• Ask: "List my VLAN interfaces."
• Claude should call mikrotik_list_vlan_interfaces with empty args and surface the JSON result.
• If the call fails with a connection error, double-check --host, the SSH user's policies, and that /ip ssh is allowed on the device.

MCP Inspector

The Inspector is the official debugging tool for MCP servers. It launches a server, connects to it, and gives you a browser UI to browse the tool catalog and invoke them by hand. No LLM involved — pure introspection.

Quick one-liner

$ npx @modelcontextprotocol/inspector \
    uvx mcp-server-mikrotik \
    --host <HOST> \
    --username <USERNAME> \
    --password <PASSWORD> \
    --port 22

Against an mcp-config.json

If you've already prepared a config file (recommended for MCPO too), install the inspector globally and point it at the file:

$ npm install -g @modelcontextprotocol/inspector
$ cp mcp-config.json.example mcp-config.json
$ nano mcp-config.json   # edit credentials
$ mcp-inspector --config mcp-config.json --server mikrotik-mcp-server

Open the URL it prints. The left pane lists every tool; clicking one opens a form for its arguments and a "Call" button. The response JSON appears beneath. Excellent for learning the schema before you wire a real client.

MCPO — REST API

MCPO ("MCP-to-OpenAPI") wraps any MCP server in a FastAPI shell. Each tool becomes a POST endpoint, and you get auto-generated OpenAPI docs at /docs. Use this for:

• ChatGPT custom actions (which expect OpenAPI URLs).
• n8n / Make / Zapier-style automation.
• curl scripts and CI jobs.

Install

# Option 1 — uvx (recommended, no install)
$ uvx mcpo --help

# Option 2 — pip
$ pip install mcpo

Config file

Create mcp-config.json in your project directory. Adjust the connection parameters (host, username, password, port) to match your setup:

{
  "mcpServers": {
    "mikrotik-mcp-server": {
      "command": "python",
      "args": [
        "src/mcp_mikrotik/server.py",
        "--password", "admin",
        "--host",     "192.168.1.1",
        "--port",     "22",
        "--username", "admin"
      ],
      "env": {}
    }
  }
}

Run the server

# With API key authentication
$ uvx mcpo --port 8000 \
    --api-key "your-secret-key" \
    --config ./mcp-config.json

# Or without auth (not recommended for production)
$ uvx mcpo --port 8000 --config ./mcp-config.json

You'll see two URLs in the startup log:

Calling a tool

Each tool is exposed at /<server-name>/<tool-name>. POST a JSON body containing the tool's arguments.

$ curl -X POST http://localhost:8000/mikrotik-mcp-server/mikrotik_list_ip_addresses \
    -H "Authorization: Bearer your-secret-key" \
    -H "Content-Type: application/json" \
    -d '{}'