Your Infrastructure Is Now Agent-Native
The NetBox Labs Platform MCP Server and new Agent Skills ensure every infrastructure management workflow is addressable by agents
On June 11, NetBox Labs announced its infrastructure intelligence platform, a unified system that lets teams and AI agents model, see, act on, and govern infrastructure with confidence. Coinciding with NetBox’s 10th anniversary, the announcement includes NetBox Data Exchange, NetBox Asset Lifecycle, NetBox Labs Platform MCP Server & Agent Skills, and NetBox Validation. Read the main blog post on this news for all the details.
AI agents have moved from chat sidekicks to operational actors. Across our customer base, teams are connecting Claude Code, ChatGPT, Cursor, and custom agent frameworks to production systems and asking them to read, write, and orchestrate real infrastructure. Last week, Cisco announced their Cloud Control agent platform, integrated with NetBox Labs. Gartner predicts that 70% of enterprises will deploy agentic AI to operate their IT infrastructure by 2029, up from less than 5% in 2025. The question has shifted. It is no longer “can the model understand my infrastructure?” It is “can my infrastructure be operated safely by an agent, and does the agent actually know what it is doing?”
Today we are answering both questions, with two announcements: the NetBox Labs Platform MCP Server, now in Public Preview, and a library of open source Agent Skills, now generally available.
One gives agents access to the NetBox Labs platform. The other gives them the knowledge to operate it correctly. Access plus knowledge, through one platform.
Why access alone was never enough
Two things became true over the past year. First, the Model Context Protocol (MCP) emerged as the industry standard for connecting agents to data and systems. Anthropic donated the protocol to the Linux Foundation’s Agentic AI Foundation in late 2025, MCP SDK downloads have surpassed 97 million per month, and the ecosystem has indexed more than 17,000 MCP servers, including the open source NetBox community MCP server which has seen wide adoption and become the “hello world” for infrastructure agents. The integration tax that defined the last decade of tooling, one custom wrapper per data source, is being replaced by a universal connector.
Second, and less discussed, is that access is necessary but not sufficient. Give an agent an API connection and it will still hallucinate fields, miss pagination, and reach for deprecated patterns, because it lacks structured knowledge of how the platform actually works. And direct write access is unsafe by default: no sandbox, no per-user attribution, no audit trail. That is why so many agent projects stall in proof-of-concept: the model can answer questions, but security and compliance will not approve ungoverned writes to production, and nobody wants to own the integration plumbing in between.
We built the agent-native platform to close both gaps at the same time.
Platform MCP Server: one interface, the entire platform
The Platform MCP Server is a fully managed MCP server that makes the NetBox Labs platform directly operable by AI agents. Our community MCP server remains a great learning tool, but the Platform MCP server unlocks real autonomous operations at scale. We run the deployment; you connect your MCP client and go.
At Public Preview, the Platform MCP server ships with nearly 100 tools spanning the full NetBox operational surface (reads, writes, bulk operations, IPAM allocation, cable tracing, GraphQL, config rendering, and scripts), plus deep integration with a broad array of NetBox Labs products and features including Branching, Changes, Custom Objects, and the newly announced NetBox Validation, and more. Dynamic model discovery indexes every object type at startup, including any installed plugins, so the tool surface grows automatically as your platform evolves. There is no new wrapper to build, no schema to relearn.
Code Mode is one of the capabilities we are most excited about. Instead of an agent making six sequential tool calls to complete a multi-step task, it writes a single block of Python that runs in a sandboxed executor. In our internal benchmarking, this approach cut context substantially and can cut tool call round trips by 75% or more on multi-step tasks with capable models.
The Platform MCP Server works with any MCP-compatible client, including Claude Code, Claude Desktop, ChatGPT, Cursor, VS Code, and custom agent frameworks, so the work you do once carries across whatever tools your teams choose. It’s how Cisco integrates their Cloud Control agents with NetBox Labs, and now every agent can interact with the NetBox Labs platform.
Agent Skills: agents that know NetBox
Connecting an agent is half the job. Teaching it to operate the platform like a senior engineer is the other half. That is what Agent Skills do.
Agent Skills is an open library, published at github.com/netboxlabs/skills under Apache 2.0 and installable in Claude Code, Cursor, and any compatible client. The skills encode the conventions a NetBox expert already knows: API integration patterns, data modeling, branching workflows, scripts, config templates, plugin development, migration, branching, and commercial capabilities including, Changes, NetBox Custom Objects, NetBox Validation, NetBox Asset Lifecycle Management, NetBox Data Exchange, and much more. A hub skill routes an agent to the right specialist for any task, and progressive disclosure keeps context efficient: core instructions stay concise, and detailed references load only when needed.
Without skills, agents guess at conventions and invent API fields. With them, agents follow the same best practices a senior engineer would. And because the library is open and actively maintained, new skills ship as new capabilities ship.
Safe by design, not by bolt-on
For most teams, the blocker to agent rollout is not capability, it’s governance. We designed the platform so security and compliance can say yes.
Every tool call carries per-user authentication with full NetBox RBAC, attributed to the human who authorized the agent. Agent writes can be directed through Branching: make agents work in isolated branches, propose diffs, self-correct issues found with pre-change analysis by NetBox Validation, and merge only through human review. Code Mode is sandboxed, with no file I/O, no imports, and execution and loop limits. Read-only mode, GraphQL cost analysis, and tier-enforced tool registration round out the layered safety model, so capabilities a customer does not have simply do not exist in their client. The result turns an agent rollout from a risk conversation into a control conversation.
A platform that compounds
Because the MCP surface tracks the NetBox Labs platform, connecting once means your agents grow more capable as the platform does: each new infrastructure intelligence capability is reachable through the same connection, with no new integration work to take on. Validation tools are live at Public Preview today, with Assurance, Discovery and other products coming online through the same connection soon. And all of this arrives over the system of record your teams already run.
Getting started
During Public Preview, the NetBox Labs Platform MCP Server is included in every NetBox Cloud plan, with capabilities mapped to existing plan tiers:
- Free: read-only tools and model introspection
- Starter: adds writes and GraphQL
- Professional and Premium: everything listed above plus Code Mode, advanced NetBox features, and tools for the entire NetBox Labs portfolio.
Agent Skills are open source and free for the community, forever.
NetBox is the system of record the infrastructure world already runs on, trusted by tens of thousands of organizations and built by hundreds of contributors over a decade in production.
Making it agent-native is the natural next step. Connect your MCP client, install the skills from GitHub, and put your agents to work.
