Connect ChatGPT or Codex

Give OpenAI's assistants read-only access to your XFA device data. Use the ChatGPT steps for the ChatGPT app, or the Codex steps for the Codex CLI and IDE extension.

ChatGPT

XFA is published in the ChatGPT app directory, which is the quickest way to add it.

Add from the directory

Open the XFA app in ChatGPT and add it.
Click Connect, sign in to XFA, and approve the one organization ChatGPT may read.
In a chat, turn XFA on from the + menu.

Add as a custom connector instead

Custom MCP connectors are a paid-plan feature (Plus, Pro, Business, Enterprise, or Edu) and are added through Developer mode.

On a personal plan (Plus or Pro)

In ChatGPT on the web, open Settings → Apps & Connectors, then Advanced settings, and turn on Developer mode.
Back on Apps & Connectors, click Create, give it a name (for example, XFA), and under Connection paste the MCP server URL https://mcp.xfa.tech/mcp. Create it.
Open the XFA connector, click Connect, sign in to XFA, and approve the one organization ChatGPT may read.
In a chat, turn XFA on from the + menu.

On Business or Enterprise

A workspace admin first allows Developer mode for the workspace (Settings → Connected data / Connectors, under the admin console).
Each member turns on Developer mode as above, clicks Create, and adds the URL https://mcp.xfa.tech/mcp.
Open the XFA connector, click Connect, sign in to XFA, and approve the one organization ChatGPT may read.

If you cannot find Developer mode, your plan or workspace does not allow custom connectors yet. OpenAI documents the current steps in Developer mode and MCP apps in ChatGPT.

Codex

The Codex CLI and IDE extension read the same configuration.

Add XFA as a remote MCP server:

codex mcp add xfa --url https://mcp.xfa.tech/mcp

Sign in with OAuth:
```
codex mcp login xfa
```
Codex opens your browser. Sign in to XFA and approve the one organization Codex may read.
Start Codex. Once the server is connected, it lists the XFA tools.

Prefer to edit the config yourself? Add this to ~/.codex/config.toml, then run codex mcp login xfa:

[mcp_servers.xfa]
url = "https://mcp.xfa.tech/mcp"

OAuth login works only with remote (streamable HTTP) MCP servers, which is what XFA runs, so no token or client id is needed. codex mcp logout xfa clears the local sign-in.

Check it worked

Ask ChatGPT or Codex "How many devices do we have?" A correct answer for your organization means the connection is live.

Revoke access

Open Connected apps in the XFA dashboard to see every connected assistant and revoke any of them at any time.