Act on intent with your agents or ours: Platform MCP is now generally available
AI agents are ready to take on real infrastructure work, from building out a new site to keeping the system of record accurate. What holds most teams back is trust. An agent that can change that record needs the same guardrails as the engineers who work in it, and without them, security won’t sign off and pilots never reach production data.
The NetBox Labs Platform MCP Server gives agents those guardrails, and it is generally available today on the SaaS edition of the NetBox Labs platform. Claude, ChatGPT, Cursor, and other Model Context Protocol (MCP) clients can use it to read your system of record and, where write access is enabled, update it under the permissions your team already sets. NetBox Labs hosts the MCP server, so there is nothing to install or operate.
We’re also announcing three new pre-built NetBox Agents in public preview: the Data Quality Steward, Change Review Agent, and Lifecycle & Version Compliance Agent.
Bring your own agent: NetBox Labs Platform MCP
Since the NetBox Labs Platform MCP entered public preview, it has given MCP clients one endpoint for the platform, and teams don’t have to build or maintain a connector. Its tools cover search, reads, writes, bulk operations, branches, and every capability of NetBox including IP address management (IPAM) and cable tracing. It discovers the object types in your instance, including installed plugins, which lets an agent work with your data model as you’ve extended it.
An engineer can ask an agent to find the planned devices at a site, check their interfaces against the device-type templates, allocate management IPs, and stage the changes in a branch. The NetBox Labs Platform MCP docs cover setup for Claude Code, Claude Desktop, ChatGPT, Cursor, and VS Code, and any client that supports MCP over streamable HTTP with an authorization header can connect.
The open source NetBox MCP server and the Agent Skills that teach agents the NetBox data model stay open source under Apache 2.0, and we keep investing in both. The open source server is deliberately simple, read-only, and self-hosted – a great way to start experimenting with agents. Platform MCP adds a fully managed endpoint with write tools for teams putting agents into production.
Or use ours: NetBox Agents
The three new agents join the Reporting, IPAM, and Troubleshooting agents from the August public preview.
The Data Quality Steward learns the naming patterns and required fields your team already uses, proposes them as standards, and stages fixes as one change request per rule, so a reviewer can accept or reject each rule on its own. Anything it can’t know, such as a serial number, goes to a person’s to-do list.
Reviewers start each change request from a first pass by the Change Review Agent, which checks it against change-board rules your team writes in plain language and flags where the diff doesn’t match its description. The agent comments and requests changes, and leaves approving and merging to people.
The Lifecycle & Version Compliance Agent reports end-of-support exposure and version compliance against the software versions you’ve standardized on, cites the source of every date, and plans upgrade waves that start with the devices that have the fewest documented dependents. This agent addresses an urgent use case we’re hearing in the market, accelerating patching to remediate risk from a surge in vulnerabilities discovered by Mythos-class models.
Teams can also describe their own agent in plain language and run it on a schedule, triggered by events, or on demand. Because agents keep working between conversations, routine requests stop queueing behind the team. NetBox Labs provides model access through its own platform, so teams don’t need to connect a client or manage a model key.
The same infrastructure guardrails for people and agents
Each person connects to Platform MCP with their own API token, so an agent can do only what that person is allowed to do. A connection can also be opened read-only, which hides every write tool and rejects any write call.
NetBox Agents run as a dedicated account your team scopes, so the same permissions decide what they can reach. Any agent can be set to read-only, and for everything else your team sets each type of action to Always allow, Needs approval, or Blocked. Where NetBox Branching and NetBox Change Management are enabled, agents stage their changes in a branch, and nothing merges until a change request is approved. Every agent action is logged, along with the approval when one was required.
NetBox Labs Platform MCP: Availability and access
The NetBox Labs Platform MCP is generally available on the SaaS edition, and NetBox Agents are in public preview. Contact your account team to get access. Both are coming to the self-managed edition in the future.
Not a NetBox Labs customer yet? Request a demo and we’ll walk through Platform MCP and NetBox Agents alongside the rest of the platform.
