Meeting prep
Ask ChatGPT who you are meeting and get what they said, what they want and what is still open.
Connect your email, calendar, calls, Slack and CRM once. Add one MCP server to ChatGPT and sign in. Every answer about a customer comes from the whole record, with a source on every fact.
The same chat you use today. The difference is what ChatGPT knows when you ask.
Meeting prep
Ask ChatGPT who you are meeting and get what they said, what they want and what is still open.
Pipeline reads
Which deals moved, which stalled and why, from the conversations rather than the stage field.
Customer voice
Objections, requests and decisions from Pulse, quoted as the customer said them.
Follow-ups you approve
ChatGPT drafts the email from what was agreed. It waits for a person before it goes out.
Win and loss
Why deals were won or lost, built from what actually happened, with the source of each reason.
Memory that learns
What ChatGPT learns in a chat is written back to the record, with you as its source.
You may already have Gmail, a CRM or a note-taker connected to ChatGPT, and it helps. But each connector gives it one slice. The CRM knows the stage. The call knows what was said. Slack knows what the team decided after.
The value isn’t in connecting more tools. It’s in joining them: knowing that an address in Gmail, a name on a call and a contact in the CRM are one person, and keeping what each of them said with where it came from. That is what customer memory is.
01
Connect your tools
Sign in to Northstar and connect Gmail, your calendar, your note-taker and your CRM. Northstar builds one record per customer from them and keeps it current as conversations happen.
02
Add the MCP server in ChatGPT
In Northstar, open Home and choose Connect, then ChatGPT, and copy the server address. On ChatGPT on the web, open Plugins and choose Add custom MCP server. Paste the address.
03
Sign in and ask
Choose OAuth, create it as a plugin, then sign in and authorize your workspace. In a chat, type @, pick Northstar and ask ChatGPT about a customer.
Memory reaches ChatGPT over MCP, the Model Context Protocol, the open standard for connecting AI to outside data. ChatGPT doesn’t get raw exports. It gets one resolved record per customer, scoped to you.
Compare
Individual connections
Customer memory
Who a person is
A different address, name or contact in each connector
One person, resolved across every tool
Cross-tool questions
ChatGPT stitches fragments together on every question
One call returns the whole account
Sources
Whatever each connector returns
Every fact keeps its source and history
What it may see
Whatever each connection can read
Only what its person may see
Actions
Up to each prompt
The ones that matter wait for a person
Setup
One connector per tool, kept up by you
Connect your tools once, point at one server
Other agents
Set up again for each one
Claude, ChatGPT, Grok and your own agents share it
Where do I add it in ChatGPT?
On ChatGPT on the web, under Plugins, as a custom MCP server. Northstar shows the exact steps and the server address under Connect on its Home page.
Does it work with Codex?
Yes. Run npx @opennous/cli@latest init in your project. It signs you in, creates a workspace key and registers the MCP server with Codex. Claude Code and Cursor connect the same way.
What can ChatGPT see?
Only what you may see. ChatGPT connects as you, so private records stay private and an agent never reads more than its person can. Every answer carries its source.
Can ChatGPT change anything?
ChatGPT can write what it learns back to memory, and that change is kept with you as its source. Actions that matter, such as sending an email, wait for a person to approve them unless your team has allowed them to run on their own.
Which plan do I need?
Free covers it: up to three people and 50,000 records, with the agent harness over MCP. Team adds unlimited people, the built-in note-taker, Slack and WhatsApp sync and two-way CRM sync.
Can I self-host it?
Yes. Northstar is open source and runs on your own servers with Docker. Point ChatGPT at your own server’s address instead of the hosted one.
Connect your tools, add one MCP server, and ask ChatGPT about a customer you know well.