A small Model Context Protocol (MCP) server built with FastMCP that provides an interface to a NetBox instance. This repository exposes NetBox resources (sites, site-groups, devices, etc.) as MCP tools and runs a streaming HTTP transport so clients can connect and receive streamed responses from the MCP server. - FastMCP: The MCP tool registry and runtime. Tools are defined in app.py and decorated
Add this skill
npx mdskills install simonpainter/netbox-mcpWell-documented NetBox MCP server with clear setup instructions and useful query tools
A small Model Context Protocol (MCP) server built with FastMCP that provides an interface to a NetBox instance.
This repository exposes NetBox resources (sites, site-groups, devices, etc.) as MCP tools and runs a streaming HTTP transport so clients can connect and receive streamed responses from the MCP server.
app.py and decorated with @mcp.tool.app.py that wraps httpx.AsyncClient to call the NetBox API.mcp.run(transport="http", ...) which runs an HTTP server that supports a streaming/chunked response transport. This is useful for clients that want to consume events or long-running responses incrementally rather than waiting for the entire result.Conceptually the flow is:
search_sites or get_site_group_details).app.py, which may in turn call NetBox via NetBoxClient.Note: The low-level HTTP path/shape is provided by the FastMCP runtime. Clients that wish to connect should use a compatible MCP client or an HTTP client that supports reading chunked / streaming responses.
Set these before running the server:
NETBOX_URL - Base URL to your NetBox instance (default: https://netbox.example.com).NETBOX_TOKEN - NetBox API token with read permissions.MCP_PORT - Port for the FastMCP HTTP transport. Defaults to 8000 if not set. If the value is not a valid integer, the server will exit with an error.Example (macOS / zsh):
export NETBOX_URL="https://netbox.example.com"
export NETBOX_TOKEN="0123456789abcdef0123456789abcdef01234567"
export MCP_PORT=8000
Start the MCP server with Python:
python3 app.py
The server will bind to 0.0.0.0 on the configured port and serve the FastMCP HTTP transport.
app.py registers MCP tools for NetBox resources. Examples included in this repository:
search_sites — search of dcim/sites/ (supports partial/case-insensitive name queries)get_site_details — site lookupsearch_site_groups — search of dcim/site-groups/ (supports name__ic)get_site_group_details — single site-group lookup by idYou can add more tools following the repository conventions: each NetBox resource has a search_ and get__details tool.
python3 -m py_compile app.py
curl --no-buffer or a custom MCP client) to connect to the FastMCP HTTP transport and invoke tools.Because the exact HTTP request shape (paths, headers and payload) is defined by the FastMCP runtime and clients, prefer using an existing MCP client library where available. If you're building a custom client, make sure it supports reading chunked responses so streamed results can be consumed incrementally.
If you add tools, follow the repo conventions documented in .github/copilot-instructions.md: each resource should have a search_ and a get_ tool with descriptive docstrings and consistent parameter mapping.
This project is provided as-is.
Install via CLI
npx mdskills install simonpainter/netbox-mcpNetBox MCP Server is a free, open-source AI agent skill. A small Model Context Protocol (MCP) server built with FastMCP that provides an interface to a NetBox instance. This repository exposes NetBox resources (sites, site-groups, devices, etc.) as MCP tools and runs a streaming HTTP transport so clients can connect and receive streamed responses from the MCP server. - FastMCP: The MCP tool registry and runtime. Tools are defined in app.py and decorated
Install NetBox MCP Server with a single command:
npx mdskills install simonpainter/netbox-mcpThis downloads the skill files into your project and your AI agent picks them up automatically.
NetBox MCP Server works with Claude Code, Claude Desktop, Cursor, Vscode Copilot, Windsurf, Continue Dev, Gemini Cli, Amp, Roo Code, Goose. Skills use the open SKILL.md format which is compatible with any AI coding agent that reads markdown instructions.