# Connect ChatGPT to Your Business Data

> Connect ChatGPT to your own business data through a custom connector authenticated with OAuth 2.1 and PKCE, so it searches before it answers.

Canonical page: https://www.greatarrow.ai/guides/connect-chatgpt-to-your-business-data · updated 2026-09-15

## The problem: ChatGPT doesn't know anything about your business

Out of the box, ChatGPT is a capable general assistant with no idea what's in your inbox, your team's Slack channels, your CRM, or your GitHub repos. Ask it something that depends on your actual data — "what did we decide about the Q2 roadmap" or "summarize the last email from this client" — and it has nothing to work with beyond whatever you paste into the chat window. Every question that touches your business ends up requiring you to first go find the answer yourself, then paste it in for ChatGPT to reformat or reason over.

Custom connectors exist to close that gap: instead of a generic chat window, ChatGPT gets a live channel to a specific server that can search real content on your behalf, mid-conversation, without you copying anything in by hand. The setup is a one-time OAuth connection; the payoff is every future ChatGPT conversation being able to reach your actual data instead of working from what you happen to remember to paste.

This guide covers connecting ChatGPT to GreatArrow.ai as a custom connector, and what changes once it can search your workspace directly.

## How it works

GreatArrow.ai exposes an MCP (Model Context Protocol) server at `https://www.greatarrow.ai/api/mcp`. ChatGPT connects to it as a web-based client using [OAuth 2.1 with dynamic client registration and PKCE](https://developers.openai.com/api/docs/mcp) — you register the connector once and approve access through ChatGPT's own consent screen, rather than generating and pasting a token.

Once connected, ChatGPT can call tools against your workspace: searching memory, documents, and any integrations you've connected (Gmail, Drive, Slack, Notion, GitHub, and others). The MCP server's `initialize` response carries an `instructions` field specifically written to push search-before-answer behavior into connecting clients — so a connector-aware ChatGPT session leans toward checking your workspace before answering a question that references your specifics, rather than guessing or asking you to paste the answer in.

## Setting it up

1. **Create your account.** Sign up at [/sign-up](/sign-up) — no credit card required for the 14-day trial.
2. **Connect your business data sources first (optional but recommended).** Anything ChatGPT should be able to search — Gmail, Drive, Slack, Notion, GitHub, and other supported integrations — needs to be connected to your account before ChatGPT can search it. You can also add these later; ChatGPT picks up new sources as soon as they're connected.
3. **Open the install page and choose ChatGPT.** Go to [/install](/install), signed in, and select ChatGPT. This is a guided OAuth path.
4. **Enable developer mode in ChatGPT on the web** under **Settings → Security and login** or **Settings → Apps → Advanced settings**. Then open **Settings → Apps → Create** and add a custom MCP app (or **Plugins → +** if your account still shows the older interface) pointing at the endpoint the install page shows you. On a managed Business, Enterprise, or Edu workspace, ask an OpenAI owner/admin to enable developer mode and connector access if these controls are hidden.
5. **Complete the OAuth consent screen.** ChatGPT walks you through sign-in and asks you to approve the connection — pick the Great Arrow workspace you want this OpenAI workspace to use, then approve.
6. **Confirm the connection.** The install page can detect a successful OAuth connection automatically. If detection doesn't fire, a manual "I did this" confirmation is available.
7. **Test it.** Ask ChatGPT something that depends on a source you connected — a recent email, a document, a Slack thread — and confirm the answer reflects real content rather than a generic response.

## What "search before answer" actually looks like

Once the connector is live, a ChatGPT conversation that touches your specifics doesn't stay generic. Ask about a project, a decision, or a person your team has discussed, and ChatGPT is nudged to check your connected workspace before responding — pulling from memory, documents, or synced integration content rather than answering from general knowledge alone. This doesn't replace your judgment on what to connect: ChatGPT can only search what's actually in the workspace, so the more of your real data sources you connect, the more questions it can answer directly instead of asking you to supply the context yourself.

The range of "business data" here is broad by design. Beyond Google Workspace (Gmail, Drive, Calendar, Docs, Sheets, Slides) and Microsoft 365 (Outlook, Teams, OneDrive, SharePoint), the same account can connect developer tools (GitHub, GitLab, Jira, Asana, Linear), communication tools (Slack, Discord, Notion, Confluence), and CRM data (GoHighLevel). Each is connected independently in your integration settings, and once connected, ChatGPT's connector can search across all of them from the same conversation — you're not switching contexts to ask about an email versus a ticket versus a Slack thread.

## Revoking access later

Disconnecting ChatGPT is a client-level action, separate from the data sources themselves. Removing the custom connector from ChatGPT's settings ends that client's access to your workspace without touching the underlying integrations — Gmail, Slack, and everything else stay connected and searchable from any other client you've connected. If you want to stop a specific data source from being searchable at all, that's handled separately, by disconnecting that integration from your account rather than by removing the ChatGPT connector.

## Where to go next

If Notion is one of your team's primary knowledge sources, [ChatGPT MCP Server for Notion](/mcp-servers/chatgpt-mcp-server-for-notion) covers connecting that integration specifically. If you also use Claude and want both clients working from the same connected data, [Share Memory Between Claude and ChatGPT](/guides/share-memory-between-claude-and-chatgpt) covers running both side by side.

## Questions

### What plan does ChatGPT need to add a custom connector?

Availability depends on OpenAI's developer-mode rollout and your workspace policy: full MCP including write actions is rolling out to ChatGPT Business, Enterprise, and Edu, and eligible Pro users may have read/fetch access in developer mode. Managed workspaces may also require an OpenAI admin to enable developer mode under Workspace settings → Permissions & Roles. ChatGPT's free tier doesn't support custom MCP apps.

### Does ChatGPT need a separate integration setup for each data source?

No. You connect ChatGPT once, as a custom connector pointed at one MCP endpoint. Individual data sources — Gmail, Drive, Slack, Notion, and others — are connected separately in your account, and once connected, ChatGPT can search all of them through that same single connector.

### How does ChatGPT authenticate, and is a token involved?

ChatGPT authenticates with OAuth 2.1, dynamic client registration, and PKCE — you approve access through an in-app consent screen, the same shape as any other "sign in with X" flow. There's no API token to copy or paste for this client.

### Does ChatGPT search my data automatically, or do I have to ask it to?

The connector's initialize response includes instructions that push ChatGPT toward searching before answering questions that reference your specifics — projects, decisions, people, files. You don't have to prompt it to search explicitly for that behavior to kick in, though results still depend on what's actually connected and stored.

### Is my data used to train OpenAI's models?

Great Arrow never supplies your workspace content to any third party for model training, and it's not used for the platform's own model training without your separately-given consent. What OpenAI does with your own ChatGPT conversations is governed by your ChatGPT data-controls settings with OpenAI, not by Great Arrow.
