netbox-mcp-server
Gives AI assistants read (and optionally write) access to a NetBox instance, covering DCIM, IPAM, circuits, virtualization, tenancy, and more.
README
netbox-mcp-server
A Model Context Protocol server that gives an AI
assistant read (and optionally write) access to your NetBox
instance — DCIM, IPAM, circuits, virtualization, tenancy, power, and the
netbox-inventory plugin.
Written in TypeScript on the official @modelcontextprotocol/sdk. Runs locally over
stdio as a subprocess of any MCP-aware client: Claude Desktop, Claude Code, Codex CLI,
Cursor, Continue.
Installing this? Paste this into ChatGPT, Claude, or any assistant that can browse and run commands:
Read https://raw.githubusercontent.com/zenixsolutions/netbox-mcp-server/main/AGENTS.md and follow it to install the NetBox MCP server on my Mac.
AGENTS.mdis a step-by-step runbook written for an AI assistant to execute without guessing — dependency checks, build, client config, and troubleshooting. Humans can use the Quick start below instead.
Quick start
# 1. Clone
mkdir -p ~/mcp-servers && cd ~/mcp-servers
git clone https://github.com/zenixsolutions/netbox-mcp-server.git
cd netbox-mcp-server
# 2. Check dependencies, install, build, smoke-test
./scripts/install.sh
install.sh verifies Homebrew, git, and Node >= 18, runs npm ci && npm run build,
confirms the server answers a tools/list request, and prints a ready-to-paste client
config with the correct absolute paths filled in. It never runs sudo.
Then get a NetBox API token (NetBox → your user menu → API Tokens → Add a token) and add the server to your client:
{
"mcpServers": {
"netbox": {
"command": "/opt/homebrew/bin/node",
"args": ["/Users/YOU/mcp-servers/netbox-mcp-server/dist/index.js"],
"env": {
"NETBOX_URL": "https://netbox.yourcompany.com",
"NETBOX_TOKEN": "your-api-token",
"NETBOX_READONLY": "1"
}
}
}
}
Claude Desktop reads ~/Library/Application Support/Claude/claude_desktop_config.json.
Fully quit (Cmd-Q) and reopen it afterwards. Other clients:
AGENTS.md § 6.
If you have NETBOX_URL and NETBOX_TOKEN exported, ./scripts/install.sh --write-claude-config will merge the entry in for you (backing up your existing config
first).
What it exposes
446 tools across 89 NetBox resources. Every resource gets netbox_list_<plural>,
netbox_get_<singular>, netbox_create_<singular>, netbox_update_<singular> (PATCH
semantics), and netbox_delete_<singular> — plus netbox_global_search, which fans a
fuzzy query out across resources in parallel.
| Group | Tools | Covers |
|---|---|---|
search |
1 | cross-resource fuzzy lookup |
dcim |
40 | sites, locations, racks, manufacturers, device types, device roles, platforms, devices, interfaces, cables |
dcim_org |
20 | regions, site groups, rack roles, rack types, rack reservations |
dcim_components |
76 | modules, module/device bays, console + front + rear ports, MAC addresses, inventory items, and their templates |
ipam |
32 | prefixes, IP addresses, VLANs, VLAN groups, VRFs, aggregates, IP ranges, roles |
ipam_org |
16 | RIRs, ASNs, ASN ranges, route targets |
ipam_services |
24 | services, service templates, FHRP groups, VLAN translation policies + rules |
inventory |
28 | netbox-inventory plugin: assets, suppliers, purchases, deliveries |
power |
24 | power panels, feeds, ports, outlets, and templates |
tenancy |
24 | tenants, tenant groups, contacts, contact roles/groups/assignments |
virtualization |
28 | clusters, cluster types/groups, VMs, VM interfaces, virtual disks |
circuits |
44 | providers, provider accounts/networks, circuits, terminations, circuit groups, virtual circuits |
deletes |
89 | netbox_delete_* for every resource |
All list tools support:
- rich per-resource filters (site, status, role, tenant, VRF, …)
- universal filters —
q(fuzzy text),tag[],created_after,created_before - pagination via
limit/offset, withhas_moreandnext_offsetin the response response_format—markdown(default) orjson- automatic truncation at 25,000 characters, with guidance to paginate
Configuration
| Variable | Required | Default | Meaning |
|---|---|---|---|
NETBOX_URL |
yes | — | Base URL of your NetBox, e.g. https://netbox.corp.com. Omit /api — the server appends it. A trailing / or /api is stripped for you. |
NETBOX_TOKEN |
yes | — | NetBox API token. |
NETBOX_INSECURE |
no | off | 1/true/yes skips TLS certificate verification. For on-prem NetBox behind an internal CA. |
NETBOX_READONLY |
no | off | 1/true/yes registers only list, get, and search tools (179 total). Create, update, and delete tools are not registered at all. |
NETBOX_TOOL_GROUPS |
no | all | Comma-separated allowlist of groups from the table above, e.g. search,dcim,ipam. Unknown names are ignored with a warning on stderr. |
Both NETBOX_READONLY and NETBOX_TOOL_GROUPS default to off — with neither set, the
full 446-tool surface is registered, exactly as before these options existed.
Suggested rollout profiles
# Most people: read-only, full visibility -> 179 tools
NETBOX_READONLY=1
# Read-only, trimmed for clients that struggle
# with large tool counts -> 49 tools
NETBOX_READONLY=1
NETBOX_TOOL_GROUPS=search,dcim,ipam,tenancy
# Engineers who document as they work: write,
# but no delete tools exist at all -> 233 tools
NETBOX_TOOL_GROUPS=search,dcim,dcim_org,dcim_components,ipam,ipam_org,power,tenancy
# Administrators: everything, deletes included -> 446 tools
# (set nothing)
NETBOX_TOOL_GROUPS is an explicit allowlist — anything you do not name is omitted,
including deletes. Name deletes deliberately if you want it.
Verifying an install
# Does the server start and register tools?
printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"cli","version":"1"}}}' \
'{"jsonrpc":"2.0","method":"notifications/initialized"}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
| NETBOX_URL=https://netbox.corp.com NETBOX_TOKEN=... node dist/index.js \
| tail -1 | python3 -c 'import sys,json; print(len(json.load(sys.stdin)["result"]["tools"]))'
# Do the credentials work?
curl -sS -H "Authorization: Token $NETBOX_TOKEN" \
"$NETBOX_URL/api/dcim/sites/?limit=1" | head -c 200
# Interactive exploration — starts a local web server and runs until Ctrl-C
npx @modelcontextprotocol/inspector node dist/index.js
Then ask your assistant: "Using the netbox tools, list the first 5 sites."
./scripts/install.sh --check re-runs just the dependency checks at any time.
Troubleshooting
The most common failure by far: spawn node ENOENT in a GUI client. Claude Desktop
and ChatGPT are launched from Finder and never source your ~/.zshrc, so a node
installed by nvm/fnm/asdf/Volta is invisible to them. Use the absolute path from
which node in the config, not the bare string "node". install.sh detects this and
fills in the absolute path for you.
Claude Desktop logs each server separately:
tail -f ~/Library/Logs/Claude/mcp-server-netbox.log
Full table of symptoms and fixes: AGENTS.md § 8.
Development
npm run dev # tsx watch src/index.ts
npm run build # tsc -> dist/
npm run clean
src/
index.ts entry point; registers tool groups
constants.ts character limits, env var names
config.ts env parsing / validation
gating.ts NETBOX_READONLY + NETBOX_TOOL_GROUPS
client.ts axios-based NetBox client
errors.ts friendly API error formatting
formatting.ts markdown rendering + pagination payload
registrars.ts list/get/create/update/delete tool factories
schemas/common.ts shared Zod schemas
tools/ per-domain resource registrations
scripts/
install.sh macOS dependency check, build, config generator
Adding a resource means adding a descriptor + schemas in the relevant tools/*.ts file;
the registrars generate the five tools. New groups must also be added to
ALL_TOOL_GROUPS in src/gating.ts and wired in src/index.ts.
Security notes
- The token is read only from
NETBOX_TOKEN. It is never logged or echoed in a tool response. - The NetBox token is the real permission boundary. Issue read-only tokens in NetBox
itself for anyone who does not need write access;
NETBOX_READONLYis a convenience layer that keeps the tools out of the model's context, not an enforcement mechanism. - Deletes cascade and cannot be undone. NetBox removes dependent objects: deleting a
device removes its interfaces, power ports, and assigned IPs; deleting a site can
remove its racks, devices, and prefixes.
deletetools carrydestructiveHint: trueand a description that insists on confirming first — but the safe posture for most users is to leave them unregistered. NETBOX_INSECURE=1disables TLS certificate validation entirely. Prefer installing your corporate root CA into the system keychain.- Never commit a token.
.envis gitignored; client configs live outside the repo.
Contributing
Issues and pull requests welcome — see CONTRIBUTING.md.
Security vulnerabilities should be reported privately, not as public issues. See SECURITY.md, which also covers how to operate this safely (token scoping, cascading deletes, and prompt-injection risk with write access).
Disclaimer
This is an independent, community-maintained project. It is not affiliated with, endorsed by, or supported by NetBox Labs or the NetBox open-source project. "NetBox" is a trademark of its respective owner.
Provided as-is under the MIT license. You are responsible for what an AI assistant does with the credentials you give it — read the security notes above before granting write access to a production NetBox instance.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.