Getting started

By the end of this you will have a read-only API admin on a FortiGate, an MCP client that can reach it, and an answer back from the appliance confirming the whole path works.

Roughly ten minutes, most of it on the firewall.

What you need

A FortiGate running FortiOS 6.4 or later, administrative access to it, and a machine that can reach its management interface. You also need uv installed, which is what uvx comes from.

1. Create a read-only admin profile

   The API user needs a profile, and the built-in ones grant more than this server can use. Make one that grants read and nothing else.

   In the web UI: System > Admin Profiles > Create New. Name it readonly-api and set every permission group you care about to Read. The groups that matter are System, Network, Firewall, and — if the appliance has radios — WiFi & Switch Controller.

   From the CLI, which is faster and less error-prone:

   config system accprofile
       edit "readonly-api"
           set secfabgrp read
           set ftviewgrp read
           set authgrp read
           set sysgrp read
           set netgrp read
           set loggrp read
           set fwgrp read
           set vpngrp read
           set utmgrp read
           set wifi read
       next
   end

2. Create the REST API admin

   System > Administrators > Create New > REST API Admin.

   Give it a username, set Administrator Profile to readonly-api, leave PKI Group disabled, and set Trusted Hosts to the address of the machine that will run the server.

   Trusted hosts is the control that matters

   A FortiGate API token is a bearer credential: anything holding it is that admin. Trusted hosts is what makes a leaked token useless from anywhere but one address. Do not skip it, and do not set it to 0.0.0.0/0 to get past an error.

   The CLI form, with the trusted host set to the machine you will run from:

   config system api-user
       edit "mcp-readonly"
           set accprofile "readonly-api"
           set vdom "root"
           config trusthost
               edit 1
                   set ipv4-trusthost 192.0.2.10 255.255.255.255
               next
           end
       next
   end

3. Generate the token

   Saving the API admin in the web UI shows the token once, in a dialog. There is no way to retrieve it afterwards — you generate a new one instead — so copy it somewhere before closing.

   From the CLI:

   execute api-user generate-key mcp-readonly

4. Check it from the command line first

   Before involving an MCP client, confirm the credential works. This is the step that separates the token is wrong from the client configuration is wrong, and doing it now saves a confusing half hour later.

   curl -sk "https://fgt.example.com/api/v2/cmdb/system/global?access_token=YOUR_TOKEN" \
     | head -c 200

   A JSON body with "status": "success" means the token, the trusted host, and the profile are all correct. HTTP 401 means the token is wrong or the source address is not in the trusted hosts list.

5. Add the server to your MCP client

   Not released yet

   The first release has not been published, so uvx mcfortigate will not resolve until it is. Until then, run it from a checkout: point the client’s command at uv run --directory /path/to/mcfortigate mcfortigate instead.

   Claude Code

   claude mcp add fortigate \
     --env FORTIGATE_HOST=fgt.example.com \
     --env FORTIGATE_TOKEN=your-token \
     -- uvx mcfortigate

   Claude Desktop

   In claude_desktop_config.json:

   {
     "mcpServers": {
       "fortigate": {
         "command": "uvx",
         "args": ["mcfortigate"],
         "env": {
           "FORTIGATE_HOST": "fgt.example.com",
           "FORTIGATE_TOKEN": "your-token"
         }
       }
     }
   }

   Any MCP client

   The server runs over stdio. The command is uvx mcfortigate, and configuration arrives entirely through the environment:

   FORTIGATE_HOST=fgt.example.com FORTIGATE_TOKEN=your-token uvx mcfortigate

   uvx fetches and runs the package without installing it anywhere permanent, so there is no virtualenv to manage and upgrades are automatic.

6. Ask it something

   Restart the client so it picks up the new server, then ask:

   What FortiGates can you reach?

   The model calls list_targets and reports one appliance with its URL, VDOM, and auth mode. No credentials come back — they never leave the server process.

   Then ask for the appliance itself:

   What model and firmware is it running?

   That is get_system_status, and a serial number in the answer means the whole path works, because the serial comes from a part of the FortiOS response that a broken read would have thrown away.

If something did not work

A client that shows the server as failed usually has a configuration error rather than a connection problem. The server prints its version and its configured targets to stderr at startup, and clients that surface server logs will show it:

mcfortigate v2026.9.11 | targets: fgt.example.com

If that line says targets: none configured, the environment variables did not reach the process, which is a client configuration problem rather than a FortiGate one. Troubleshooting covers the rest.

Next

Three questions worth asking first walks through what the tools are actually good at, using a real appliance.