NetBox MCP Server — 179 AI Tools | Claude, ChatGPT & More

NetBox MCP ServerNetBox MCP server with 179 ready-to-use AI tools. Connect NetBox to Claude, ChatGPT, Gemini, or any MCP-compatible AI assistant. Cloud MCP connector — no install, no Docker, works from any browser.Category: Developer Tools. 179 tools available. Authentication: API Key.Query your entire data center infrastructure with plain EnglishWhy NetBox + AI?NetBox holds your source of truth for devices, IPs, racks, and circuits -- but navigating its UI for quick answers is slow. Let AI query your infrastructure data instantly so you can focus on operations.Example prompts you can send your AI about NetBox

Ask: "Which rack has available space in the NYC data center?"
AI does: AI queries racks at the NYC site, checks elevation data, and shows racks with available units.

Ask: "Find all IPs in the 10.0.0.0/8 range that are assigned to servers"
AI does: AI searches IP addresses within the prefix, filters by assignment, and lists devices with their IPs.

Ask: "List all circuits from Lumen Technologies with their terminations"
AI does: AI fetches circuits by provider, includes termination details, and shows connected sites.

Ask: "Show me all devices with role "router" that are offline"
AI does: AI queries devices filtered by role and status, presenting a clear list for your NOC team.

Featured NetBox Tools

list-devices: List devices in NetBox inventory with optional filters. Returns device name, type, site, rack position, status, and primary IP. Use this as your FIRST tool when the user asks about devices, servers, routers, switches, or firewalls. FILTER RULES: All string filters use SLUGS (lowercase-hyphenated), NOT display names. WRONG: site="New York DC1" — CORRECT: site="new-york-dc1". WRONG: role="Core Router" — CORRECT: role="core-router". WRONG: manufacturer="CommScope" — CORRECT: manufacturer="commscope". You MUST call list-sites, list-device-roles, or list-manufacturers first to find valid slugs before filtering. Do NOT guess slugs. STATUS VALUES: "active" (in production), "planned" (not yet deployed), "staged" (ready to deploy), "failed" (broken), "offline" (powered down), "decommissioning" (being removed), "inventory" (in storage). EXAMPLES: All active devices at a site: site="nyc-dc1", status="active". All Cisco routers: manufacturer="cisco", role="router". Find device by name: q="core-rtr-01". WHEN TO USE: User asks "show me all devices", "what servers are at site X?", "list routers", "what devices are offline?", or any device inventory question. For executive summaries (counts by role, by site, by manufacturer), use graphql-query instead — it can fetch all devices with their roles and sites in one query.

search-ip: Search for a specific IP address and see which device/interface it is assigned to. Returns the IP record with assignment details, status, VRF, and DNS name. You can search with or without CIDR mask — "10.0.1.15" and "10.0.1.15/32" both work. WHEN TO USE: User asks "who has IP 10.0.1.15?", "what device is 192.168.1.1 on?", "is IP X assigned?", or mentions a specific IP address. This is faster than list-ip-addresses when you know the exact IP. For browsing a subnet, use list-ip-addresses with parent filter instead.

list-prefixes: List network prefixes (subnets) in NetBox IPAM. Returns prefix in CIDR, status, utilization percentage, VLAN assignment, site, VRF, and role. KEY FILTERS — "within" vs "contains": • within="10.0.0.0/8" — returns all subnets INSIDE 10.0.0.0/8 (e.g., 10.1.0.0/24, 10.2.0.0/16). • contains="10.1.2.3" — returns the prefix that CONTAINS this IP (e.g., 10.1.2.0/24). STATUS VALUES: "container" (parent prefix, subdivided into children), "active" (in use for IP assignment), "reserved" (allocated but not active), "deprecated" (being retired). EXAMPLES: All subnets in 10.0.0.0/8: within="10.0.0.0/8". Which subnet contains IP 10.1.2.3: contains="10.1.2.3". Subnets at a site: site="nyc-dc1". WHEN TO USE: User asks "what subnets do we have?", "what subnet is IP X in?", "show subnets at site X", "what is the utilization of our prefixes?", or any subnet/prefix question. For utilization overview across all prefixes, use get-available-ips per prefix or graphql-query for bulk prefix data.

list-circuits: List network circuits (WAN links, internet connections, MPLS VPNs, point-to-point links). Returns circuit ID (cid), provider, type, status, commit rate, and termination sites. FILTER RULES: provider and type filters use SLUGS. You MUST call list-circuit-providers or list-circuit-types first to find valid slugs. Do NOT guess provider slugs. STATUS VALUES: "planned" (ordered, not installed), "provisioning" (being set up), "active" (live), "offline" (down), "deprovisioning" (being removed), "decommissioned" (cancelled). EXAMPLES: Circuits from a provider: provider="level3". Active circuits at a site: site="nyc-dc1", status="active". Find by circuit ID: cid="CKT-001234". WHEN TO USE: User asks "what circuits do we have?", "show WAN links at site X", "what ISP connections are active?", or any WAN/circuit question. For executive summaries (circuit counts by provider, by status, by site), use graphql-query: { circuit_list { cid status provider { name } type { name } } }.

get-rack-elevation: Get a visual representation of all devices mounted in a rack, showing which rack units (U positions) are occupied and which are free. Returns each U position with the device name (if occupied) or empty status. WHEN TO USE: User asks "what is in rack X?", "show rack layout", "which U positions are available?", "rack elevation diagram".

graphql-query: Execute a custom GraphQL query against NetBox. Use this for complex queries that combine multiple object types, need specific field selections, or require data that no single REST tool provides. NetBox GraphQL is READ-ONLY — you cannot create, update, or delete objects with it. TYPE NAMING: All GraphQL types use the "Type" suffix: DeviceType, InterfaceType, SiteType, RackType, IPAddressType, PrefixType, VLANType, CableType, VMInterfaceType, CircuitType, TenantType. CRITICAL: In inline fragments you MUST use the Type suffix. WRONG: "... on Interface { }". CORRECT: "... on InterfaceType { name device { name } }". FILTER SYNTAX: Two kinds of filters exist — STRING filters use nested lookup objects, ENUM filters take a direct value. WRONG: device_list(name: ["router-01"]). WRONG: device_list(site: $site). STRING FILTERS use nested lookups: device_list(filters: {name: {exact: "router-01"}}). device_list(filters: {site: {name: {contains: "nyc"}}}). STRING FILTER OPERATORS (StrFilterLookup): exact, contains, i_contains, starts_with, ends_with, in_list, is_null, regex. ENUM FILTERS take a DIRECT value — NO nested {exact:} wrapper. CORRECT: device_list(filters: {status: STATUS_ACTIVE}). WRONG: device_list(filters: {status: {exact: ACTIVE}}). WRONG: device_list(filters: {status: ACTIVE}). ENUM VALUES ALWAYS have a prefix matching the field context. Device/Site/IP/Prefix/VLAN/VM status: STATUS_ACTIVE, STATUS_PLANNED, STATUS_STAGED, STATUS_FAILED, STATUS_OFFLINE, STATUS_INVENTORY, STATUS_DECOMMISSIONING, STATUS_RESERVED, STATUS_DEPRECATED, STATUS_DHCP, STATUS_SLAAC, STATUS_CONTAINER, STATUS_STAGING, STATUS_RETIRED. Circuit status (exception — no prefix): ACTIVE, PLANNED, PROVISIONING, OFFLINE, DEPROVISIONING, DECOMMISSIONED. IP address role: ROLE_LOOPBACK, ROLE_SECONDARY, ROLE_ANYCAST, ROLE_VIP, ROLE_VRRP, ROLE_HSRP, ROLE_GLBP, ROLE_CARP. IP address family: FAMILY_4, FAMILY_6. Device face: FACE_FRONT, FACE_REAR. Interface type: TYPE_VIRTUAL, TYPE_LAG, TYPE_1000BASE_T, TYPE_10GBASE_SR, etc. (132 values — use introspection if unsure). Interface mode: MODE_ACCESS, MODE_TAGGED, MODE_TAGGED_ALL. Combining filters: device_list(filters: {status: STATUS_ACTIVE, site: {name: {contains: "nyc"}}}) — enum and string filters side by side. PAGINATION: device_list(pagination: {limit: 50, offset: 0}). SCALAR FIELDS (CRITICAL) — In GraphQL, choice fields return PLAIN STRINGS, NOT objects. NEVER use { label } or { value } subfields on them. WRONG: face { label }. WRONG: status { value }. WRONG: type { label }. CORRECT: face (returns "front"/"rear"/null). CORRECT: status (returns "active"/"planned"/etc.). CORRECT: type (returns "1000base-t"/"virtual"/etc.). Full list of scalar choice fields by type — DeviceType: status, face, airflow. InterfaceType: type, mode, speed, duplex, rf_role, rf_channel, poe_mode, poe_type. IPAddressType/PrefixType/VLANType/SiteType/VirtualMachineType: status. RackType: status, airflow, form_factor, weight_unit, outer_unit. CircuitType: status, distance_unit. CableType: status, type, color, length_unit. VMInterfaceType: mode. DeviceTypeType: airflow, subdevice_role, weight_unit. These fields ARE objects (with subfields like name, slug): site, rack, role, device_type, manufacturer, tenant, platform, primary_ip4, primary_ip6, vlan, vrf, cluster. REVERSE RELATIONS — field names vary by type, do NOT assume "devices" exists everywhere: DeviceRoleType has "devices" and "virtual_machines". DeviceTypeType has "instances" (NOT "devices"). SiteType has "devices", "racks", "prefixes", "vlans", "circuit_terminations". RackType has "devices". ManufacturerType has "device_types" (NOT "devices"). COUNT FIELDS only exist on DeviceType (the device object): interface_count, console_port_count, console_server_port_count, power_port_count, power_outlet_count, front_port_count, rear_port_count, device_bay_count, module_bay_count, inventory_item_count. Other types (RackType, DeviceRoleType, etc.) do NOT have count fields — query nested lists and count client-side. UNION TYPES: ip_address_list returns "assigned_object" which is a union of InterfaceType | VMInterfaceType | FHRPGroupType. You MUST use inline fragments: assigned_object { ... on InterfaceType { name device { name } } ... on VMInterfaceType { name virtual_machine { name } } ... on FHRPGroupType { name } }. COMMON ROOT TYPES: device_list, interface_list, ip_address_list, prefix_list, vlan_list, site_list, rack_list, circuit_list, virtual_machine_list, cable_list, tenant_list, device_role_list, manufacturer_list, device_type_list. EXECUTIVE QUERY EXAMPLES (all verified against live schema): Devices grouped by type: { device_type_list { model manufacturer { name } instances { id } } } — count instances client-side. Devices by role with sites: { device_role_list { name devices { id name site { name } status } } }. All devices with primary IPs: { device_list { name status site { name } device_type { model manufacturer { name } } primary_ip4 { address } } }. Active devices at a site: { device_list(filters: {status: STATUS_ACTIVE, site: {name: {exact: "my-site"}}}) { name role { name } device_type { model } } }. IPs assigned to interfaces: { ip_address_list(filters: {status: STATUS_ACTIVE}) { address dns_name assigned_object { ... on InterfaceType { name device { name } } } } }. Interface counts per device: { device_list { name interface_count console_port_count power_port_count } }. Rack capacity overview: { rack_list { name site { name } u_height devices { id } } } — count devices array client-side. WHEN TO USE: User asks executive summary questions spanning multiple types (e.g., "how many devices per role?", "list all IPs with their assignments"), or when you need to minimize API calls by fetching related data in one query. WHEN NOT TO USE: For simple single-type queries, prefer the dedicated REST tools (list-devices, list-ip-addresses, etc.) as they are simpler.

Frequently Asked Questions about NetBoxHow do I connect NetBox to Claude or ChatGPT?Create a free LearnToSpeak.ai account, add your NetBox credentials in the dashboard, and paste the generated OAuth URL into Claude.ai's custom connectors or ChatGPT's developer apps. The integration runs fully in the cloud — no desktop app, no CLI, no local install.Does it cost anything to connect NetBox?The NetBox integration is free during the LearnToSpeak.ai public beta. You only need a paid plan from Anthropic or OpenAI (Claude Pro/Max/Team or ChatGPT Plus/Pro/Team) since custom connectors require it on the AI provider side.Which AI assistants support the NetBox connector?Claude.ai (Pro, Max, Team, Enterprise) and ChatGPT (Plus, Pro, Team) both support the NetBox integration via the MCP (Model Context Protocol) standard. The same credentials work across both platforms — add the connector once, use it everywhere.How many NetBox tools does my AI get access to?Your AI gets 179 ready-to-use tools for NetBox, covering the most common read and write operations. Each tool is a discrete action the AI can call on demand — no multi-step prompting, no screen scraping, no workarounds.Is it safe to give an AI access to my NetBox account?Yes. Your NetBox credentials are encrypted at rest with AES-256-GCM and only decrypted in memory when the AI makes a request. LearnToSpeak.ai uses OAuth 2.1 with mandatory PKCE for the AI-to-platform connection, and you can revoke access instantly from the dashboard. We never store or log the actual tool responses returned to the AI.Why use LearnToSpeak.ai instead of running an MCP server locally?Cloud MCP servers work from any browser on any device — no Node.js install, no Docker, no local ports, no maintenance. Configure NetBox once in the dashboard and your AI connects from claude.ai or chat.openai.com directly. That's the entire value proposition: web AI assistants don't read your filesystem, they read our servers.What is the NetBox MCP server?The NetBox MCP server is a cloud-hosted integration that gives AI assistants direct access to NetBox through 179 tools via the Model Context Protocol (MCP) standard. It works with Claude, ChatGPT, Gemini, and any MCP-compatible AI — no local installation or Docker required.How do I connect NetBox to Claude via MCP?Sign up at LearnToSpeak.ai, add your NetBox credentials, and copy the generated MCP URL into Claude.ai's custom connectors (Settings → Developer → Custom Connectors). The cloud MCP server handles the connection — no CLI, no local server, no Docker.How do I connect NetBox to ChatGPT via MCP?Sign up at LearnToSpeak.ai, add your NetBox credentials, and copy the generated MCP URL into ChatGPT's developer apps (Settings → Apps → Advanced → Developer mode). Works with any ChatGPT Plus, Pro, or Team account.What can AI do with NetBox through MCP?NetBox offers 179 MCP tools covering Infrastructure resource modeling for network and data center management. Ask your AI to perform any of these actions in natural language — it calls the right tool automatically.What is MCP (Model Context Protocol)?MCP is an open standard that lets AI assistants like Claude, ChatGPT, and Gemini connect directly to external apps and services. LearnToSpeak.ai hosts cloud MCP servers so you don't need to install anything locally — your AI connects from any browser.Related MCP Servers

GitHub MCP Server — 44 tools

GitLab MCP Server — 90 tools

Cisco CPNR MCP Server — 36 tools

Grafana MCP Server — 50 tools

Graylog MCP Server — 12 tools

Browse all 34 MCP servers