NetPilot

NetPilot in ChatGPT, Claude and Meta Muse

NetPilot adds a network lab to the AI assistant you already use. Connect it once, then ask ChatGPT, Claude or Muse to design a topology, deploy it on real network operating systems, run commands on the devices and hand you the verified configs for your change window. The lab runs on your own NetPilot lab VM, so the devices, the configs and the designs are yours.

What you need

  • A NetPilot account on Pro, Max, Team or Signature. A Team seat or a Signature seat in an organization counts, which covers Enterprise customers on NetPilot cloud. A dedicated or on-prem Enterprise deployment has its own endpoint: ask your NetPilot contact for it. On Free or Plus the tools answer with a note that the feature starts at Pro.
  • A lab VM. You do not have to create it first: the first time you ask the assistant to start the lab, NetPilot creates the VM for your account, the same way the Start button in the web app does.
  • An assistant that supports remote MCP servers: ChatGPT, Claude.ai, Claude Code or Meta Muse. Each client's own plan and workspace policy decide whether you can add one.

Public URL

https://agent.netpilot.io/mcp

Paste this when a client asks for the server URL. Sign-in uses your NetPilot account, and the assistant asks for one permission: NetPilot agent access. No API key is needed.

How to connect

When NetPilot is listed in your assistant's directory, connect it from there. Until then, or on a client without a listing, add it by URL. Both paths end at the same sign-in.

Claude.ai

Available on Free, Pro, Max, Team and Enterprise Claude plans. A Free account can hold one custom connector, so remove another one first if the Add option is blocked.

  1. 1Open Customize, then Connectors. If NetPilot is listed, click Connect and sign in with your NetPilot account.
  2. 2To add it by URL, click + Add, then Add custom connector. Name it NetPilot, paste https://agent.netpilot.io/mcp and click Continue.
  3. 3Claude detects NetPilot's sign-in. Keep the detected authentication settings, choose Sign in now, leave the OAuth client choice at its default, add no headers or API keys, then click Add and sign in with your NetPilot account.
  4. 4In a chat, open the + menu, choose Connectors and switch NetPilot on. If you chose Sign in when needed instead, Claude asks you to sign in the first time it uses a tool.
  5. 5On Team and Enterprise, an owner adds NetPilot under Organization settings, then Connectors, and each member connects from Customize.

Claude Code: claude mcp add --transport http --scope user netpilot https://agent.netpilot.io/mcp then /mcp to sign in.

ChatGPT

Use ChatGPT on the web for these steps. Which ChatGPT plans can add a custom MCP server with write tools is set by OpenAI, see OpenAI's custom MCP server guide. On a workspace plan, permissions and security restrictions apply too, so ask your workspace admin if the option is missing there.

  1. 1Open Plugins in ChatGPT. If NetPilot is listed, install it and sign in with your NetPilot account.
  2. 2To add it by URL, click +, then Add custom MCP server. Name it NetPilot, paste https://agent.netpilot.io/mcp, choose OAuth as the authentication, read and accept the warning about custom servers, then choose Create as a plugin.
  3. 3Review the tools ChatGPT discovered. Then open your personal plugins (chatgpt.com/plugins, Personal tab), open NetPilot and select the plus button to install it.
  4. 4Start a Work chat, type @ and pick NetPilot. Sign in with your NetPilot account when ChatGPT asks.

Meta Muse

Connectors live under Settings in Muse. Meta offers Muse to people in the United States and Canada who are 18 and over, so the steps below apply there.

  1. 1Open Settings, then Connectors. Tap Connect next to NetPilot, read the information, tap Continue and sign in with your NetPilot account.
  2. 2If NetPilot is not listed yet, ask Muse to create a custom connector and give it https://agent.netpilot.io/mcp.
  3. 3Disconnect at any time under Settings, then Connectors.

What the assistant can do with NetPilot

Eighteen tools in three groups. The assistant picks them for you. You ask for the outcome.

Your lab VM

Start, stop and check the lab VM that runs your devices.

  • Start the VM. On first use NetPilot creates it for you.
  • Stop the VM and see the lab credits the session used.
  • Check status: running or stopped, time until the idle stop, vendor images on the VM and the labs deployed on it. The assistant reads this first and picks the next step from it.

Design and Cisco Packet Tracer

Turn a description into a topology and configs, and hand them to you in the format you work in.

  • Design a topology from plain English and read it back.
  • Generate the device configs for each vendor in the design and read them back.
  • Export the design as a Cisco Packet Tracer .pkt file. The download link works for 60 minutes.
  • Vendor guide: how each supported network OS is configured, from Nokia SR Linux and FRRouting to Cisco IOL, Arista cEOS and Juniper cRPD, so the assistant writes the right syntax the first time.

Deploy and operate

Run the lab on real network operating systems and work on the devices.

  • Push the topology and the configs to the VM and deploy the lab with containerlab.
  • Run show and configuration commands on any device, and drive interactive dialogs such as a password or a confirmation prompt.
  • Run shell commands on the VM, start long jobs in the background and read their output later.
  • The result is a set of verified configs and show output you carry into your change ticket or your cutover plan.

Permissions, limits and errors

  • Permission: one sign-in with your NetPilot account grants NetPilot agent access to the connected assistant. No API key, and the assistant never sees your NetPilot password.
  • Read and write: checking status, reading a design or a config, the vendor guide and reading a job's output are read-only. Claude runs them without a prompt unless your organization sets NetPilot or a tool to Needs approval, and Muse asks once on first use and then offers Always allow. Starting or stopping the VM, deploying a lab, pushing configs and running commands on devices change state, and clients that confirm destructive actions ask you first.
  • Side effects: a deploy or a device command runs on your lab VM only. Starting the VM spends lab credits until it stops.
  • Limits per account: 60 lab calls a minute, 20 VM start or stop calls a minute, 30 VM starts an hour, 30 design calls a minute, and two lab calls running at the same time. A single call runs for at most 16 minutes, so long jobs run in the background and the assistant reads their output afterwards.
  • Errors: every error names the cause and the next step, for example that the VM is stopped and needs starting, that it is still starting, that the lab credits for the month are used up, or that a feature starts at Pro. A VM that is briefly unreachable answers with a retry in 30 seconds.

Lab credits and the idle timeout

A lab started from an assistant uses the same lab credits as a lab started in the web app: Pro includes 50 a month, Team 60 per seat, Max 150, and Signature covers lab usage under fair use for your agreed scope. A session is charged when the VM stops. The VM stops itself when nothing has used it for 1 hour on Pro and Team, or 2 hours on Max and Signature. Lab commands count as activity, status checks do not. A stopped VM keeps its labs for 14 days on Pro and Team, and for as long as the account exists on Max and Signature. Start the VM again before that period ends and the labs are back where you left them. The pricing page has the plan details.

Questions about NetPilot in an assistant

NetPilot keeps the topology designs and generated configs you create, plus the state of your lab VM, in your NetPilot account. It receives only the requests the assistant sends to its tools, such as a topology description or a device command, and returns the result. The rest of the conversation stays with the assistant and is covered by that provider's privacy policy. The NetPilot privacy policy lists the retention periods.
Remove it in the assistant's own settings: Customize > Connectors in Claude, the Plugins page in ChatGPT, Settings > Connectors in Muse. The assistant loses access within a minute. Your NetPilot account, labs and designs stay as they are, and you can connect again at any time.
Pro, Max, Team or Signature, including a Team seat or a Signature seat in an organization. On Free or Plus you can connect and sign in, and each tool answers with a note that the feature is included from Pro. The plans are listed on the pricing page.
No. The assistant runs its own model, billed by that provider. NetPilot charges lab credits for the time your lab VM runs, exactly as it does for a lab started in the web app.
Yes. Your account has one lab VM, and every assistant and the web app work on it. A lab deployed from Claude shows up in the web app, and a lab started in the app is visible to ChatGPT or Muse. Two assistants writing a design at the same time overwrite each other, so finish one before starting the next.
A cold start of the lab VM can take a few minutes, longer than some assistants wait for one answer. Ask the assistant to check the VM status again. Nothing is lost: the VM finishes starting on its own, and the assistant continues from there.

Keep reading

Documentation

Use cases, supported devices, integrations and getting started.

Plans

What each plan includes, with lab credits and idle timeouts.

Integrations

NetBox, Nautobot, Nornir and the systems NetPilot's own agent connects to.

Support

How to reach us and what to include in your message.

Build the next lab from the chat you already have open

Create a NetPilot account, pick Pro or above, and connect NetPilot in your assistant.

Create an account

Stuck on a step? Support