AI Explained

Connectors: Give the Claude and ChatGPT Brain Your Business

Claude and ChatGPT are the sharpest analysts most teams will ever have access to. Most give them nothing to work with. Connectors fix that.

Max Doyle

8 Nov 2026

•

min read

•

Updated 9 Nov 2026

Claude and ChatGPT have a brain that can read a contract or rebuild a forecast in minutes, but it can't see your business. A connector fixes that. It's a secure link between the assistant and an app you already use, such as Gmail, Google Drive or your CRM. Claude calls them connectors, ChatGPT calls them apps, and both run on the open Model Context Protocol (MCP). Connect the brain to your data and its answers stop being generic.

Most teams use that brain as a very clever text box. Someone exports a report, pastes it in, copies the answer out and pastes it somewhere else. The model does the thinking. A person does the carrying, and the brain only ever sees what someone remembered to paste.

Think of it as a hire. A brilliant analyst with no email account, no access to the shared drive and no CRM login would be useless by Friday. That's how most companies run their AI. The intelligence isn't the bottleneck any more. Access is.

You've hired the smartest analyst you'll ever have. Then you didn't give it a login.

Same brain, different answer

You askWhich clients should I be worried about this month?

Without connectorsIt can see: your question

“Watch for late payments, falling engagement and stalled projects. Check your CRM for deals that haven’t moved and your accounts for overdue invoices.”

A checklist you could have written yourself.

With CRM, inbox and accounts connectedHubSpotGmailXero

Client A. No reply to your last three emails, and the renewal is due in 26 days.
Client B. Invoice 46 days overdue, the second late payment this year.
Client C. Deal stuck at proposal for three weeks, since their CFO joined the thread.

A list you can act on before lunch.

Illustrative example, not client data.

What is a connector in Claude?

A connector gives Claude's brain eyes and hands inside another app: a set of tools it can use there, with permission to use them as you. Connect Google Drive and Claude can search and read your files. Connect Linear and it can create an issue when you ask it to.

Setting one up takes a minute. In Claude on the web or the desktop app, open the + menu, choose Connectors, then Manage connectors, pick the app from the directory and sign in to it. Claude never sees your password: the other app grants it a permission you can revoke at any time. From then on Claude reaches for the connector when a question needs it, or you switch it on yourself.

Three things are worth knowing from the start.

It sees what you can see. A connector inherits your access in the other app. If you can't open a folder, channel or record, Claude can't either.
It can act, not only read. Many connectors can create, update, send or delete. You decide which of those Claude may do, tool by tool.
The directory does most of the work. Asana, Notion, Linear, Atlassian, Sanity and HubSpot are among the apps in Anthropic's directory. For anything that isn't listed, you can add a custom connector by pasting in a server address.

Directory connectors are available on every Claude plan, Free included, and Free users can add one custom connector. On Team and Enterprise plans an owner switches a connector on for the organisation first, then each person signs in with their own account. Some connectors are interactive and open a live board, dashboard or design right inside the chat.

What is a connector in ChatGPT?

The same idea under a different name, or three. OpenAI called them connectors until 17 December 2025, when it renamed them apps. On 9 July 2026 it moved the app directory into a plugin directory, where a plugin can bundle one or more apps with skills and templates. Existing connections kept working through both changes. People still search for "ChatGPT connectors", so we'll keep using the word.

Underneath, it's the same brain-meets-your-data idea. You connect Gmail, Drive, SharePoint or your CRM, sign in, and ChatGPT can search it, use it as a source in deep research and, where the app supports it, take actions. OpenAI's Apps SDK is built on MCP, the protocol behind Claude's connectors. Anthropic created MCP and in December 2025 donated it to the Linux Foundation under a fund it co-founded with OpenAI and Block, so neither company owns the standard any more.

Claude — Checked 9 November 2026 against Anthropic's and OpenAI's help centres and pricing pages. Both change names, menus and plan details often.

		ChatGPT
What it's called	Connectors	Apps, installed through plugins
Where to find them	The Connectors directory, from the + menu	The plugin directory
On the free plan	Directory connectors, plus one custom connector	Plugins and interactive apps; deep research with apps is limited
Your own systems	Custom connector on every plan	Developer mode on Plus and Pro (beta); admins publish custom apps on Business and Enterprise
Controlling actions	Always allow, Needs approval or Blocked, tool by tool	Asks before write actions, depending on the app; Enterprise admins can switch actions off
Rolling out to a team	An owner enables it, then each person signs in	Many apps on by default in Business; admins control access
Built on	MCP	MCP, through OpenAI's Apps SDK

Which is better? For most teams, the one they already pay for. The bigger difference is which apps each directory lists, and the last post in this series compares them side by side.

Why should you use connectors?

Because a great brain with your context beats a great brain guessing. These are the four reasons we see most often.

1Answers about your business, not business in general

Ask which clients are at risk and the brain checks your CRM, inbox and accounts, instead of listing what to look out for.

2Live data, not last month's export

It reads the deal, the document or the inbox as it is when you ask. No stale file, no wrong tab, no client's details pasted into the wrong chat.

3One question across several tools

“Which deals closed last month, and did each one get an onboarding email?” needs your CRM and your inbox. Without connectors that's two exports and a spreadsheet.

4The work gets done, not described

The brain drafts the reply, files the ticket or saves the post. You review and approve instead of retyping.

To see the difference, connect your email and calendar and try one of these:

Copy these: three first questions

Find every email from [client] in the last 30 days and list anything I promised them that I haven't done yet.
Look at my calendar for next week. Which meetings have no agenda in the invite? Draft a two-line agenda request for each, but don't send them.
Find the emails where someone is waiting on a reply from me for more than three days. Rank them by who's waiting longest.

How much should you let a connector do?

Give the brain your world in stages. Let it read first, then let it draft, and only then let it act. We call this Intersect's Read, Draft, Act ladder. Every connector we set up climbs it one rung at a time, and most never need the top rung.

Intersect's Read, Draft, Act ladderClimb one rung at a time

3Act
Sends, publishes, updates or deletes. Someone else sees it, or it's hard to undo.
EmailSend the replyCRMMove the deal stageWebsitePublish the post
In ClaudeNeeds approval, or Blocked
2Draft
Creates something a person checks before it goes anywhere.
EmailDraft a replyCRMDraft a call noteWebsiteSave a draft post
In ClaudeAlways allow, or Needs approval at first
1Read
Searches, reads and summarises. Nothing changes in the other app.
EmailSummarise the threadCRMLook up the dealWebsiteList the latest posts
In ClaudeAlways allow

Most of the value sits on the first two rungs. Most of the risk sits on the third.

In Claude, each connector lists its tools, grouped into read-only and write or delete. You set each one to Always allow, Needs approval or Blocked. These settings only ever narrow what the other app lets you do; they never add access. In ChatGPT, write actions may ask for your confirmation depending on the app's permissions and what the action could change, and Enterprise admins can switch individual actions off before an app reaches staff.

How do we use connectors at Intersect?

This blog is the example we know best. Our content plan lives in a Google Sheet, and Claude reads it for each post's brief. Claude researches and writes the post, we review it as a designed page, and it saves the post to our website through the Sanity connector as a draft. The publish button stays with a person. We wrote up that set-up step by step.

The brain does the hard part. The connectors are why it can. This post went the same way: Claude read our playbook and the posts already on the site so it wouldn't repeat them, checked the facts against Anthropic's and OpenAI's help centres, and saved the result into Sanity as a draft. Nobody exported or pasted anything. The LinkedIn images that promote our posts are built in Claude and handed to Canva, where the team can edit them.

Where do connectors break?

In seven places, and each one has a fix.

It acts as you

A connector carries your access, so a misread instruction can do anything you could. That's the case for the ladder: gate or block the top rung until you've watched the lower ones work.

Content can carry instructions

A malicious server, or an email or document written to trick the model, can try to steer what the assistant does next. Both Anthropic and OpenAI warn about this prompt injection. Only connect servers from organisations you trust, and keep write actions on approval.

The brain reads what's there

Connect a CRM nobody updates and you get confident answers about stale deals. A connector doesn't clean your data; it exposes it. Fix the records the brain will lean on most before you trust what it says about them.

Listed isn't audited

Anthropic says its verification label isn't a security audit. Read what a connector asks for when you sign in, and refuse permissions it doesn't need.

More connectors, less room

Every switched-on connector adds its tools to the conversation. Anthropic suggests its On demand mode once you have ten or more, and switching off the ones a chat doesn't need also stretches your usage limit.

Personal plans may train on it

On ChatGPT Free, Go, Plus and Pro, OpenAI may use information accessed through apps to train its models if "Improve the model for everyone" is switched on. Check the model-training setting on any personal plan before you connect work accounts, or connect them on a business plan instead.

Your software may not be listed

Industry systems, older ERPs and in-house tools often aren't in either directory. A custom connector fills the gap, but its server has to be reachable over the public internet, so a system behind your VPN needs a developer to expose it safely. That's a post of its own in this series.

What should you do next?

Connect one tool your team uses every day, read-only, this week, and ask the brain something only your data can answer. Email or Drive is usually the right first rung. Ask it three questions you'd normally answer by searching yourself. If the answers save time, let it draft. Leave sending and publishing for later.

Do this today

Your first connector in 15 minutes

Pick one app your team lives in: email, Drive or your CRM.
In Claude, open the + menu, then Connectors, then Manage connectors. In ChatGPT, open the plugin directory.
Sign in, and read the permissions it asks for before you approve.
In Claude, set the write and delete tools to Needs approval.
Ask three questions you'd normally answer by searching yourself.
Note what it got wrong. Those notes become your house rules for the next rung.

The connector series

Connectors: give the Claude and ChatGPT brain your businessYou're here
How to Connect Claude to Sanity, Step by StepLive
10 connectors every Claude user should haveComing soon
10 connectors every ChatGPT user should haveComing soon
Connectors for marketing: content production, media and performanceComing soon
Connectors for e-commerce, finance, sales, legal and operationsComing soon
Your industry's software isn't in the directory. Now what?Coming soon
Claude or ChatGPT: which connects better to your stack?Coming soon

Want the brain plugged into your business properly?

We'll work out which of your systems the brain should see first, connect them, set the permissions rung by rung, write the house rules and show your team which questions are worth asking. If the system that matters most isn't in either directory, we'll build the custom connector.

Book a call

FAQs

Do connectors cost extra?

Not in Claude. Directory connectors come with every plan, Free included, though what they return counts towards your usage. In ChatGPT the plugin directory is open on every plan, but some apps and features need Plus or higher, and the other app may need a paid account of its own.

Can Claude or ChatGPT see everything in my account?

Only what your own login can see in that app, and only when the connector is in use. You can disconnect at any time from the assistant's settings or from the other app's security page.

Is a connector the same as Zapier or an API integration?

No. An integration runs a fixed workflow someone designed. A connector gives the assistant tools and lets it decide when to use them, based on what you ask. They work well together: Zapier has its own connector, which lets Claude reach apps that don't have one.

Do I need a developer?

Not for anything in the directories. You need one to build a custom connector for a system that isn't listed, or to make an internal tool reachable safely.

Do connectors work on my phone?

Yes, once they're added. Set them up on the web or desktop app first, then they're there in the mobile app the next time you sign in.