AI connections

How AI connects to business systems.

The basic connection path

1

AI tool

The chatbot, agent, workflow, or application requesting access.

2

Connection

The authentication or integration method that creates the technical link.

3

Business system

The application, data source, file repository, database, or service being reached.

4

Allowed access

The specific data, functions, and actions available through the connection.

Connection methods at a glance

Connection typePrimary purposeTypical identityCan support actions?
OAuthDelegate access to another application.User or applicationYes, if granted
API keyAuthenticate direct API requests.Application or accountYes, if the API allows it
Service accountProvide persistent non-human access.Dedicated service identityYes
MCPExpose tools and functions to an AI agent.Varies by serverYes
Webhook / automationTrigger downstream workflows from events.Application or workflow identityYes
Browser extensionUse information and actions available in the browser.Signed-in userSometimes
RAG / file indexingRetrieve information from approved knowledge sources.User, application, or index serviceUsually read-oriented
CLI / SDKInteract with systems programmatically.User, service, or applicationYes

Common AI connection types

These methods solve different integration needs. Organizations may use several of them at the same time.

Delegated access

OAuth

Allows an AI application to access another application on behalf of a signed-in user or organization.

Common useEmail, calendars, cloud files, CRM systems, collaboration tools.
Access is controlled byRequested scopes, user permissions, admin consent, and tenant policy.
Direct application access

API key

A credential that allows the AI tool or application to call a system's API directly.

Common useSaaS applications, data services, internal APIs, analytics platforms.
Access is controlled byThe API endpoints, key scopes, associated account, and permissions.
Non-human identity

Service account

A dedicated account used by an AI agent or background process instead of an individual employee.

Common useAutomations, scheduled processes, enterprise integrations, background agents.
Access is controlled byAssigned roles, account privileges, credential lifetime, and environment scope.
Tool interface

MCP

A standardized way to expose approved tools, data sources, and functions to an AI application or agent.

Common useAgent tools, enterprise data access, application actions, internal systems.
Access is controlled byThe tools exposed by the server, authentication, underlying system permissions, and tool configuration.
Event-driven

Webhook or automation

Sends an event or data payload from one system to another when a defined trigger occurs.

Common useWorkflow triggers, notifications, form submissions, status changes, task creation.
Access is controlled byThe trigger, destination, downstream workflow permissions, and actions configured after the event.
User-session access

Browser extension

Allows AI to interact with information and functions available through the user's browser session.

Common usePage summarization, form assistance, browser automation, research, data capture.
Access is controlled byBrowser permissions, active sessions, allowed sites, page content, and extension settings.
Knowledge retrieval

RAG or file indexing

Makes approved documents or knowledge sources searchable so AI can retrieve relevant information while responding.

Common usePolicies, procedures, internal knowledge, client files, document libraries.
Access is controlled byIndexed sources, folder permissions, document permissions, partitions, and retrieval rules.
Programmatic execution

CLI or SDK

Allows custom software or an AI coding environment to interact with applications, infrastructure, or services through code.

Common useDeveloper tools, cloud administration, custom integrations, deployment workflows.
Access is controlled byRuntime identity, code permissions, credentials, environment boundaries, and available commands.

Connector support varies by AI platform

Claude logoClaude
ChatGPT logoChatGPT
Copilot logoCopilot
Perplexity logoPerplexity
Gemini logoGemini
Grok logoGrok
The same business system may be supported across several AI platforms. Availability can vary by plan, account type, region, and administrator settings. Compare connectors

What a platform compatibility badge can mean

Direct connectorBuilt into or officially offered for the AI platform.
Marketplace connectorInstalled through the platform's connector or app marketplace.
Premium data sourceLicensed data available directly to the AI platform.
Custom connectionAPI, MCP, webhook, or another integration configured separately.

Three concepts that determine actual access

The connection method explains how the link works. These three concepts explain how much access the AI actually receives.

Permissions

What can it do?

Read, create, update, send, delete, approve, execute, administer, or manage settings.

Scope

How much can it reach?

One file, one folder, one mailbox, one project, an entire department, or the full organization.

Identity

Whose access is it using?

An individual user, a service account, an application identity, or an administrative account.

Looking for a specific connector?

Use the Connection Risk Directory to search common applications and view their potential exposure assuming broad access.

Open Connection Risk Directory