Skip to content

FOUNDATIONS

APIs, MCPs & Connectors

How AI reaches your company’s tools and information.

01 / 09 APIs, MCPs & Connectors API A DEFINED REQUEST AND RESPONSE APPLICATION THE AI APPLICATION CONNECTOR AUTH · FIELDS · RETRIES PERMISSIONS DECIDE WHAT PASSES CRM SYSTEM OF RECORD MCP · ONE WAY TO DESCRIBE WHAT IS THERE THE REQUEST, AND THE DATA IT RETURNS /LEARN/APIS-MCPS-CONNECTORS

A useful AI system needs a way to reach the work. An API is an interface a service exposes. A connector is an integration that uses interfaces like that. MCP is a shared protocol that can help AI applications discover and use tools and context.

A useful way to picture it

Think of an office. The API is the service counter with a defined set of requests. A connector is the courier who knows how to use that counter. MCP gives participating AI applications and servers a common way to describe what is available. The analogy helps with the roles; each service still has its own permissions and behavior.

01API02Connector03MCP04Webhook

How the pieces actually work

API

A defined request and response

An application asks for something specific, such as a CRM record. The API defines the accepted inputs and the response. Credentials and permissions determine what the application may access.

CONNECTOR

The integration people actually use

A connector handles the practical work of reaching a system: authentication, translating fields, making requests, and handling failures. The word is used differently across products; check what a particular connector can actually read and write.

MCP

A common interface for AI applications

An MCP client communicates with a server that can expose tools, resources, and prompts. The application decides how those capabilities are presented and invoked. The protocol does not make a tool safe or grant unrestricted access.

WEBHOOK

A notification that something happened

Instead of repeatedly asking whether a record changed, a service can notify an endpoint when an event occurs. Verify the sender, handle duplicate delivery, and decide what should happen if a later step fails.

What this looks like in a business

For a lead-research workflow, the system might read a CRM record, retrieve approved company context, and draft a brief. Updating that CRM record is a separate action with its own permissions and review requirements. Start by making the read path reliable before enabling writes.

Try this before you automate

Can you name the source, the allowed action, and the person who owns access? If one is missing, the connection is not ready for unattended work.

Read the primary documentation

Guide reviewed September 2026. Vendor capabilities and permissions vary.

KEEP LEARNINGTools & SkillsContinue