Documenting Your Network with Your AI Agent: The Complete Guide

Every network has a documentation problem, and the size of the network does not change the failure mode. A homelab with a router, a switch, and three servers drifts out of date the same way a data center with two thousand devices does, because the records only stay accurate when someone remembers to update them, and people have better things to do with their attention. Devices get replaced, IPs get reassigned, cables get moved, and each unlogged change pushes the documentation further from reality. Before long the records become a historical artifact instead of an operational tool, and the team stops trusting them.

An AI agent changes that because it removes the effort from the equation: an assistant that can read your source of truth, answer questions about it, and record changes as they happen keeps documentation current almost by itself. This guide walks through the setup from an empty account to a working agent, with every click spelled out. You need about ten minutes and one network to document, whether that is a homelab rack in a closet or a floor in a colocation facility.

Step 1: Create Your Obelinf Account

Go to obelinf.com, sign in to the dashboard, and create an organization for the network you are documenting. The free Personal plan is built for homelabs and small setups, and it is ready for business when you outgrow them.

If you are documenting an enterprise network, start on the same free plan to try the workflow and upgrade to a team plan when you need more sites, devices, or members.

Your organization can start completely empty. There is no requirement to create a site before connecting the agent, because the agent can create sites, racks, and devices for you through MCP once it has a key. If you prefer to build the structure yourself in the dashboard, you can define your rack management setup and start your device inventory at your own pace, but do not feel pressured to enter everything now. The agent can help you fill in the gaps once it is connected.

Step 2: Create Your API Key

On the dashboard, open the Organization section in the sidebar and click API & MCP. Click the Create Key button, give the key a name such as “Claude Documentation Agent”, and choose a permission level:

• Read Only lets the agent query your records without changing anything, which is the safest way to start.
• Full Access allows it to create, update, and delete records, which you will want once you trust the workflow and you have the audit trail to review it.

You can also set an expiry if you want the key to rotate automatically.

Click Create, and Obelinf shows the key exactly once, together with a ready to paste MCP configuration that already contains your organization. Copy the token, it will not be displayed again. If you ever lose it, revoke it and create a new one.

Step 3: Connect Claude or Any MCP Client

With the key in hand, add the MCP server to your AI client. Claude Desktop and Claude Code accept an mcpServers block in their configuration, and Cursor exposes the same setting under its MCP settings. The configuration looks like this:

{
  "mcpServers": {
    "obelinf": {
      "url": "https://api.obelinf.com/<your-org-slug>/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

You can copy the configuration since it already contains your organization slug and your key. Restart the client, then verify the connection by asking your agent what tools it can see or to describe what is currently in your inventory. If it reports an authorization error, the key was probably truncated when copied, so paste it again and check that it belongs to the same organization as the slug in the URL.

Step 4: Start Documenting with Prompts

Your agent is now a documentation teammate. The fastest way to see it work is to hand it a real task in plain language. The prompts below cover the common cases, and they work at any scale because they ask the agent to operate on your actual records rather than on what it assumes:

Audit the device inventory and list every device that has no rack position.
Check the subnets and flag any IP addresses that are not assigned to a device.
Document the new server: a Proxmox node at 192.168.1.15 with 64 GB of RAM, connected to switch port 12.
Summarize what changed in the last seven days.

Run the audit prompts first. They cost nothing and they show you how the agent reasons about your records, including its limits. When you ask it to write, remember that every change lands in the changelog with a timestamp and field level diffs, so nothing the agent does is invisible. If it makes a mistake, correct the record and move on, and the audit trail shows exactly what happened. For a deeper look at the available tools, see AI Agent Integration.

Step 5: Guardrails for Teams and Enterprises

The same setup scales from a personal lab to a team because the controls live in the platform, not in the prompt. Keep every agent on a Read Only key until you have reviewed its diffs a few times, then promote it to Full Access deliberately. Make each key revocable and scoped to its purpose, so a key used by one agent can be revoked without affecting anyone else. For teams, role based access control and SSO decide who can create keys and what members can see, and the changelog gives you an audit trail of every agent edit with attribution.

Multi site organizations follow the same pattern, one site per location, with the agent able to work across all of them or a scoped key per site if you prefer to separate concerns. When you plan address space for any of those sites, IP address management covers subnets and the addresses inside them. The free plan includes API and MCP access, so the pattern is identical whether you are documenting one rack or a hundred.

An AI Agent That Writes into Your Source of Truth

Documentation used to be the tax you paid for running infrastructure. With an AI agent connected to a structured source of truth, it becomes a background process that keeps itself current. Obelinf gives you that foundation: a connected data model where devices, racks, IPs, and cables reference each other, a network topology view that reflects your records, and an MCP endpoint with scoped keys on every plan. Add your agent, point it at your organization, and the records start describing reality instead of history.