nautobot-mcp-server 0.2.1

Nautobot app and Model Context Protocol (MCP) server exposing Nautobot's REST and GraphQL APIs to LLMs.

nautobot-mcp-server

A Model Context Protocol server that exposes Nautobot's built-in REST and GraphQL APIs as tools for LLMs and AI assistants such as Claude Desktop, Cursor, and any other MCP-compatible client.

It can be used in two ways:

• As a Nautobot app — drop it into a Nautobot environment, configure PLUGINS_CONFIG, and run the MCP server alongside Nautobot.
• As a standalone stdio MCP server — installed in any Python environment with the nautobot-mcp-server console script, talking to a remote Nautobot over its REST API.

This release focuses on Nautobot's internal/built-in models only (DCIM, IPAM, circuits, tenancy, virtualization, extras/jobs, users, GraphQL). Third-party Nautobot apps (Design Builder, Onboarding, SSoT, Golden Config, ChatOps, …) are not yet wrapped — but the generic rest_list / rest_get tools work against their endpoints if your token has access.

Installation

pip install nautobot-mcp-server

When running inside a Nautobot environment, install the optional nautobot extra (it pins a compatible Nautobot version):

pip install "nautobot-mcp-server[nautobot]"

Optional Nautobot app (plugin) integrations

The base install exposes only Nautobot core tools. Tools for the NetworkToCode apps are optional and can be added via extras:

pip install "nautobot-mcp-server[design-builder]"   # Design Builder tools
pip install "nautobot-mcp-server[onboarding]"        # Device Onboarding tools
pip install "nautobot-mcp-server[ssot]"              # SSoT tools
pip install "nautobot-mcp-server[golden-config]"     # Golden Config tools
pip install "nautobot-mcp-server[nornir]"            # Nornir backend tools
pip install "nautobot-mcp-server[all]"               # all of the above

By default (plugins = ["auto"]) a plugin's tools are registered when its Python package is importable — so installing an extra is all you need when the MCP server shares the Nautobot environment.

For a standalone MCP server talking to a remote Nautobot, the app lives on the remote instance, not in the MCP server's venv. In that case don't rely on auto; enable the tools explicitly:

export NAUTOBOT_MCP_PLUGINS="design_builder,ssot"   # or "all"

Use the list_active_plugins tool to see what's configured, enabled, and available.

Configure as a Nautobot app

In your nautobot_config.py:

PLUGINS = ["nautobot_mcp_server"]

PLUGINS_CONFIG = {
    "nautobot_mcp_server": {
        "nautobot_url": "https://nautobot.example.com",
        "nautobot_token": "YOUR_API_TOKEN",
        "verify_ssl": True,
        # Optional
        "ca_bundle": "",
        "request_timeout": 30,
        "max_pagination_records": 5000,
        "allow_writes": False,
        "log_level": "INFO",
        # Tenant scoping (MSP multi-tenant isolation); empty = no scoping
        "tenant_scope": [],          # e.g. ["acme", "globex"]
        "tenant_group_scope": [],    # e.g. ["managed-customers"]
    }
}

Then run the MCP server alongside Nautobot:

nautobot-mcp-server

It reads the same configuration from PLUGINS_CONFIG when launched from a Nautobot environment.

Configure as a standalone MCP server

Set environment variables and run the console script:

export NAUTOBOT_URL="https://nautobot.example.com"
export NAUTOBOT_TOKEN="YOUR_API_TOKEN"
nautobot-mcp-server

All configuration knobs:

Env var	Default	Purpose
NAUTOBOT_URL	— (required)	Nautobot base URL.
NAUTOBOT_TOKEN	— (required)	Nautobot API token.
NAUTOBOT_MCP_VERIFY_SSL	true	false disables TLS verification.
NAUTOBOT_MCP_CA_BUNDLE	—	Path to a CA bundle (private CA setups).
NAUTOBOT_MCP_TIMEOUT	30	HTTP timeout in seconds.
NAUTOBOT_MCP_MAX_PAGINATION_RECORDS	5000	Hard cap when paginate=True.
NAUTOBOT_MCP_ALLOW_WRITES	false	Enable POST/PATCH/PUT/DELETE tools.
NAUTOBOT_MCP_TENANT_SCOPE	—	Comma-separated tenant names/slugs to restrict access to.
NAUTOBOT_MCP_TENANT_GROUP_SCOPE	—	Comma-separated tenant group names to restrict access to.
NAUTOBOT_MCP_PLUGINS	auto	Plugin integrations to enable.
NAUTOBOT_MCP_LOG_LEVEL	INFO	Log level (logs go to stderr).

Connection credentials stay as NAUTOBOT_URL / NAUTOBOT_TOKEN. MCP-specific knobs use the NAUTOBOT_MCP_ prefix. Legacy unprefixed names (NAUTOBOT_ALLOW_WRITES, NAUTOBOT_TENANT_SCOPE, MCP_LOG_LEVEL, …) are still accepted; the prefixed form wins when both are set.

Hooking into Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "nautobot": {
      "command": "nautobot-mcp-server",
      "env": {
        "NAUTOBOT_URL": "https://nautobot.example.com",
        "NAUTOBOT_TOKEN": "YOUR_API_TOKEN"
      }
    }
  }
}

Other transports are also supported via CLI flag:

nautobot-mcp-server --transport streamable-http
nautobot-mcp-server --transport sse

Read-only by default

Mutating operations (rest_create, rest_update, rest_delete) are blocked unless you explicitly set NAUTOBOT_MCP_ALLOW_WRITES=true (or allow_writes: true in PLUGINS_CONFIG). GraphQL queries and Nautobot Job runs are allowed regardless, since they are sandboxed by Nautobot's own permission model and the API token's scope.

Tenant scoping (multi-tenant / MSP isolation)

For MSP-style deployments where one server (and one Nautobot token) serves agents acting on behalf of different customers, you can lock the server to one or more tenants. Set tenant_scope (tenant names/slugs) and/or tenant_group_scope (tenant group names), via env vars or PLUGINS_CONFIG:

export NAUTOBOT_MCP_TENANT_SCOPE="acme"
# or multiple / by group:
export NAUTOBOT_MCP_TENANT_SCOPE="acme,globex"
export NAUTOBOT_MCP_TENANT_GROUP_SCOPE="managed-customers"

When a scope is active, enforcement happens centrally in the client, so an agent cannot cross customer boundaries regardless of the filters it passes:

• List/search on tenant-aware endpoints (devices, racks, locations, prefixes, IP addresses, VLANs, VRFs, circuits, VMs, clusters) is constrained to the in-scope tenants. Any caller-supplied tenant/tenant_id filter is dropped and replaced.
• Single-object reads are verified against the scope; out-of-scope objects (including those with no tenant) raise PermissionError.
• Creates must target an in-scope tenant. If exactly one tenant is in scope it is injected automatically; with multiple, tenant must be set explicitly.
• Updates/deletes first verify the target object is in scope, and reject reassigning an object to an out-of-scope tenant.

Non-tenant-aware endpoints (manufacturers, platforms, statuses, content-types, …) are left untouched. Run the tenant_scope_info tool to see the active scope and resolved tenant UUIDs.

Note: tenant scoping constrains the REST tools. The generic graphql_query tool and Nautobot Job runs are not tenant-filtered — rely on a tenant-restricted API token and/or per-job permissions for those paths.

Tool catalog

Discovery & metadata

• nautobot_status — Nautobot version, plugins, db backend
• list_api_roots — top-level REST categories
• list_endpoints_in(category) — sub-endpoints inside dcim, ipam, etc.
• list_content_types
• tenant_scope_info — active tenant scope (if any) and resolved tenant UUIDs
• list_active_plugins — configured/enabled/available plugin integrations

Generic REST passthrough

• rest_list(endpoint, filters, limit, offset, paginate, max_records)
• rest_get(endpoint, object_id)
• rest_create(endpoint, data) (write)
• rest_update(endpoint, object_id, data, partial) (write)
• rest_delete(endpoint, object_id) (write)

DCIM

• list_devices, get_device, get_device_interfaces, get_device_config_context, get_device_inventory
• list_locations, get_location, list_location_types
• list_racks, get_rack, get_rack_elevation
• list_device_types, list_device_roles, list_manufacturers, list_platforms
• list_cables

IPAM

• list_prefixes, get_prefix, get_available_ips, get_available_prefixes
• list_ip_addresses, get_ip_address
• list_vlans, get_vlan, list_vlan_groups
• list_vrfs, list_rirs, list_namespaces

Circuits / Tenancy / Virtualization

• list_circuits, get_circuit, list_providers, list_circuit_types, list_circuit_terminations
• list_tenants, get_tenant, list_tenant_groups
• list_virtual_machines, get_virtual_machine, get_vm_interfaces, list_clusters, list_cluster_types

Extras (jobs, statuses, tags, custom fields, …)

• list_jobs, get_job, run_job, get_job_result, list_recent_job_results, get_job_logs
• list_statuses, list_tags, list_roles
• list_custom_fields, list_relationships, list_computed_fields, list_config_contexts
• list_webhooks, list_secrets_groups

Users

• list_users, list_groups, list_object_permissions

GraphQL

• graphql_query(query, variables)
• graphql_introspect

Cross-domain search shortcuts

• search_devices, search_ip_addresses, search_prefixes, search_locations, search_circuits

Optional plugin tools

Registered only when the matching plugin is enabled (see Optional Nautobot app integrations).

• Design Builder — list_designs, list_design_deployments, list_design_jobs, run_design
• Device Onboarding — list_onboarding_jobs, onboard_device
• SSoT — list_ssot_jobs, list_ssot_syncs, list_ssot_sync_logs, run_ssot_sync
• Golden Config — list_config_compliance, compliance_summary, list_compliance_rules, list_compliance_features, list_golden_config, list_golden_config_settings, get_intended_config, list_golden_config_jobs, run_golden_config_job
• Nornir — nornir_plugin_info (backend used by Golden Config / Onboarding; device tasks run via those apps' jobs)

Development

pip install -e ".[dev]"
ruff check nautobot_mcp_server tests
pytest -v
python -m build

Roadmap

• More wrappers for popular Nautobot apps (Golden Config, ChatOps, …) using the optional-plugin pattern.
• Streaming responses for very large result sets.
• HTTP /health endpoint for the streamable-http transport.

License

MIT