How the tools work
What is MCP? Model Context Protocol explained in plain English
MCP is the protocol that lets Claude connect safely, read-only, to a business's real systems, turning a single chat into a working team of agents.
A three-layer diagram, top to bottom: a sub-agent, below it an MCP connection, below that the company's systems. Two separate lanes run between the connection and the systems. The read lane is solid and open, carrying information up from the system to the agent. The write lane is dashed and stops halfway at a gate that only opens when a person approves, before anything is written to the system.
sub-agent
A dedicated agent with its own context: a document analyst, a QA agent
MCP connector
One pipe per system. Nothing custom gets built for each one.
The systems you already have
Google Drive
CRM
Excel files
This is what the drafts-only rule means in practice: the agent can see everything, and change nothing without someone approving it.
Israel is the world's number-one user of Claude per capita, at roughly 4.9 times the global average, per Anthropic data reported by Calcalist in April 2026. That means a large share of Israeli employees already talk to Claude daily, but most companies haven't connected it to their real systems yet, which is exactly where MCP comes in.
MCP, short for Model Context Protocol, is how Claude connects to outside systems, like Google Drive, an Excel file, a CRM, or a working folder, without someone building a custom integration for each one. Instead of an employee copying data into chat by hand, the connector gives Claude direct access to the source.
The connection defaults to safe. Claude reads from a system and doesn't write to it, unless someone explicitly approves that. That's the idea behind the 'drafts only' guardrail, human-in-the-loop review of every output before it goes anywhere, not blind automation running unchecked.
A sub-agent and an MCP connector are two different, complementary things. A sub-agent is a dedicated agent with its own context, say a document analyst or a QA agent, and an MCP connector is the pipe it uses to reach an outside system. Together they make up the three waves Wider Group's course teaches: a single co-pilot, then one agent running a full task, then a system of several agents working together.
Frequently asked questions
What is MCP (Model Context Protocol) in plain English?
MCP is the protocol that lets Claude connect to a business's real systems, like a CRM, a shared drive, or a folder of files, without a custom integration built for each one. In practice, it's the difference between an employee copying data into chat by hand and an agent reaching the source directly and working from current information.
Can Claude read my company's systems safely through MCP?
Yes, by default. An MCP connection is read-only, so Claude can see what's in a system without changing it, unless someone explicitly approves a specific action. That's the 'drafts only' guardrail in practice. A human reviews every output before it goes anywhere.
What's the difference between an MCP connector and a sub-agent?
A sub-agent is a dedicated agent with its own context, like a document analyst or a QA agent. An MCP connector is the pipe that agent uses to reach an outside system, like a drive or a CRM. Wider Group's course teaches both together, as part of the three waves: co-pilot, single agent, multi-agent system.
Keep learning
Agentic AI vs. Chat
Chatting with AI and delegating to it are two different skills.
View topic