Skip to content
Connector Scout
–saved records

Guide

What are AI connectors?

An AI connector is a controlled bridge between an AI assistant and a system that holds your data. The vocabulary differs by platform — app, connector, extension, data source, MCP server — but the questions you should ask are the same in every case.

An AI connector is a controlled bridge between an AI assistant and a system that holds your data. It lets the assistant look something up, use it in an answer, and — in some cases — write something back. The bridge is deliberately narrow: it is authorised by a specific account, limited to specific operations, and usually visible in a settings screen where it can be switched off.

The vocabulary problem

Every AI platform uses different words for the same broad idea, and several have renamed their mechanisms more than once. Connector Scout uses 'AI platform' as the umbrella term and records each vendor's own terminology on the platform page, so you can search using the word you already know.

Common terms and what they usually describe
TermWhat it usually means
AppA packaged integration surfaced inside the assistant's interface.
ConnectorA first-party data connection, often to a file store or workspace tool.
ExtensionAn add-on that widens what the assistant can reach or do.
Data sourceA repository the assistant is permitted to search or index.
MCP serverA service implementing the open Model Context Protocol, which an assistant can call as a tool.
Custom connectorA connection an organisation configures itself rather than installing from a gallery.

Read, write, and act

The single most useful way to evaluate a connection is by what it can do, not by whether it exists. Connector Scout separates three access modes.

  • Read — the assistant can search and retrieve existing content.
  • Write — the assistant can create or modify content, such as drafting a document or updating a record.
  • Action — the assistant can trigger something with an external effect, such as sending a message or running a workflow.

An API is not a capability

It is tempting to assume that because a tool has a public API, an assistant can use it. That inference is wrong often enough to be dangerous. Connector Scout never records a capability on the basis of an API's existence.

Five questions to ask before enabling one

  1. Which account authorises the connection, and what can that account already see?
  2. Is the connection read-only, or can it change or send things?
  3. Does an administrator need to approve it, and does approval apply org-wide?
  4. Is content indexed and stored somewhere, or fetched on demand?
  5. How is the connection revoked, and who can revoke it?

Why this information goes stale

Connector availability, plan gating, and permission scopes change on vendor timelines, sometimes without an announcement. That is why every record in this directory carries a last-verified date and a freshness marker, and why records without a completed verification are labelled research pending rather than guessed at.

Questions

Sources