ChatGPT connectors, Claude connectors or an MCP server?

What are ChatGPT connectors and Claude connectors?

Connectors link an AI assistant to your team's tools, such as Google Drive, Gmail, Notion, Jira or GitHub. When you ask something, the assistant searches these tools, sees only what the person asking may open, and names its sources. MCP, an open standard through which assistants call other services' tools, is not the opposite of this but mostly the way underneath: you add your own connectors to ChatGPT and Claude as MCP servers.

At OpenAI, connectors are now called apps (Apps in ChatGPT). In ChatGPT Business and Enterprise, Company Knowledge answers from all connected apps, with links to the sources (Company Knowledge). At Anthropic they are called connectors. Enterprise search in Claude Team and Enterprise asks the connected tools at the moment you ask and, according to Anthropic, keeps no copy of them (enterprise search).

Scibly, our product, is itself such an MCP server, which you add to both. Here is when you need it and when the connectors are enough.

Live with every question or read in advance: what is the difference?

A connector searches when you ask. The assistant decides which tools to look in, sends each a few search words and writes the answer from the top hits. The next question starts from zero again.

A knowledge server reads beforehand. It reads each source once, recognises the people, projects and decisions in it and links them. When a document changes, it reads it again. A question then meets knowledge that is already connected.

ChatGPT sits in between: it can read some apps, such as Google Drive and SharePoint, in advance so search finds more (apps with sync).

When are the connectors of ChatGPT or Claude enough?

Often. Live search reads every document as it stands this second, keeps no copy and is included in your ChatGPT or Claude subscription. It answers questions like "What does the spec say about login?" well.

It gets weaker when the team uses several assistants, when decisions are made on calls or when you want to know what's current. Answer the three questions for your team:

• How many AI assistants does your team use?
  OneSeveral
• Where does your team decide things?
  In documents and ticketsOften on a call
• What do you ask your AI most?
  What does this document say?What's current, and why?

The connectors of ChatGPT or Claude are enough for you.

ChatGPTClaudeCursorGoogle DriveJiraNotionGoogle MeetScibly

When the whole team works in one assistant, its connectors are already there.

Why does Scibly read in advance?

Some knowledge is in no single document: who knows a topic, or which meeting changed a decision made in an email. It only shows when you read emails, meetings and documents together, and a search at the moment of the question doesn't find it. That is why Scibly reads your sources in advance.

That has a price: Scibly stores a copy of what it reads, hosted in the EU (security). On the first read it fetches the last three months, or just the last month or the last twelve if you like, and every change after that. As in the source, everyone sees in Scibly only what they may open there.

How does Scibly get into ChatGPT and Claude?

Scibly does not replace the connectors; it joins them as one more. Every assistant reaches it at the same address:

https://app.scibly.com/api/mcp

• ChatGPT. On Business and Enterprise, whoever manages your ChatGPT workspace adds Scibly as a custom MCP server and makes it available to everyone (custom MCP servers). Company Knowledge then draws on it beside your other apps.
• Claude. On Team and Enterprise, the organisation's owner adds Scibly once under Organization settings, Connectors. Then everyone connects Scibly under Customize, Connectors (custom connectors).
• Claude Code, Cursor and Copilot. There you add the same address as an MCP server.

More on how Scibly works with the two: Scibly and ChatGPT, Scibly and Claude.