Setup Guide ChatGPT

Before You Start

Ensure you have the following before proceeding:

An active Lifesight workspace. If you're evaluating Lifesight, book a demo and ask for a sandbox.
ChatGPT with Custom Connectors enabled (Team, Enterprise, or Plus)
Lifesight MCP URL: https://mcp.lifesight.io/mcp/

Steps to Connect ChatGPT Codex with Lifesight MCP

ChatGPT (Codex) uses a Personal Access Token (PAT) for authentication rather than OAuth, so there’s one extra step compared to Claude Desktop. It only takes a minute.

Generate a Personal Access Token
- Log in to the Lifesight platform.
- Navigate to the token generation URL and select your workspace.

📌
Token generation URL
https://console.lifesight.io/account/codex-tokens

Name your token (e.g. Codex or your project name) and click Generate.

Copy the generated token straight away, it won’t be shown again.
If you belong to multiple workspaces, generate a separate token for each one.

Add the MCP Server in ChatGPT Codex
- Go to Codex →Settings
- Then go to MCP Server → Add server → Streamable HTTP
- Fill in: Server name: Any label, e.g. Lifesight
- Fill in Bearer Token MCP_BEARER_TOKEN

📌
HTTP URL
https://mcp.lifesight.io/mcp

Authorization key: Enter Authorization as the key name, and add your token copied from the Lifesight earlier.
Click Save to finalize the connection.
Open a new chat and select the Lifesight MCP server to begin querying your models.

You’re connected! Start a new chat with the Lifesight MCP server selected and ask anything.

Here are some good first questions to try:
"What’s my [Quarter] budget allocation by channel?” “Which channel drove the highest iROAS last week?” “I have [ $200K] to reallocate, where should it go to maximize revenue?” “Show me how [Ad Channel 1] has been performing compared to [Ad Channel 2] over the last [N days].”

Verify Your Setup

These three prompts work in both Claude Desktop and ChatGPT. If you get structured output from your account, you’re good to go.

Prompt	Expected Outcome
List every causal model in my workspace.	Markdown table: model names, last-run date, confidence range.
What was my best-performing channel last week by iROAS?	Ranked answer with methodology footnote.
Find the methodology document for geo-lift testing.	Doc summary with link back to Lifesight.