What is MCP (Model Context Protocol)?

An AI assistant on its own only knows what you type into it. To be useful at work, it usually needs to reach your documents, calendar, tickets or databases. MCP is a common way to give it that access.

The definition

The Model Context Protocol documentation says MCP is an open-source standard for connecting AI applications to external systems. It says AI applications like Claude or ChatGPT can use MCP to connect to data sources such as local files and databases, tools such as search engines and calculators, and workflows such as specialized prompts.

The Model Context Protocol documentation says MCP provides a standardized way to connect AI applications to external systems. It likens MCP to a USB-C port.. Before USB-C, every device needed its own cable. Before a common standard, every AI app needed a custom integration for each service. With a common standard, a service builds one connector and any compatible AI app can use it.

The pieces, in plain terms

According to the Model Context Protocol docs, the MCP host is the AI application that coordinates and manages one or multiple MCP clients. According to the Model Context Protocol docs, an MCP client maintains a connection to an MCP server. According to the Model Context Protocol docs, an MCP server is a program that provides context to MCP clients. According to the Model Context Protocol docs, local MCP servers using the STDIO transport typically serve a single MCP client.

• The AI application (client): the tool you use, like Claude Code or an assistant with connectors.
• The server: a small program that exposes something, such as your Google Drive, a project tracker or a database, in a form the AI can use.
• The protocol: the shared rules for how they talk.

You typically do not build these yourself. You add a ready-made connector and grant permission.

What it enables, from the official examples

The MCP site lists examples such as agents accessing your Google Calendar and Notion, Claude Code generating an entire web app from a Figma design, and enterprise chatbots connecting to multiple databases so people can analyze data through chat. Anthropic’s Claude Code documentation similarly says Claude Code can read design docs in Google Drive, update tickets in Jira and pull data from Slack.

Why it matters if you are not a developer

MCP is the reason AI tools are moving from “paste text in a box” to “act inside your systems.” It is part of what makes AI agents practical: an agent needs hands, and connectors are its hands. If your assistant can read the project folder and post a summary to Slack without you copy-pasting, a connector is probably behind it.

What Anthropic and researchers say

• The Anthropic announcement of MCP lists pre-built MCP servers for Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer.
• An arXiv study by Xinyi Hou and colleagues defines the MCP server lifecycle in four phases: creation, deployment, operation, and maintenance.
• The same arXiv study builds a threat taxonomy covering 16 distinct threat scenarios across four major attacker types.

Risks to understand

1. Access: a connector gives the AI whatever access you grant. Grant the minimum.
2. Trust: a connector is software from someone. Use ones from sources you trust, and check with IT.
3. Untrusted content: if an AI reads a document that contains hidden instructions, it may be manipulated into acting on them. The OWASP LLM01:2025 Prompt Injection entry says these inputs can affect the model even if they are imperceptible to humans. The MCP documentation includes a security section for this reason.
4. Data flow: connected data still goes to the AI provider when the assistant processes it. Anthropic’s privacy page says the chat data it may use for model improvement does not include raw content from connectors such as Google Drive, including remote and local MCP servers, though data may be included if it is directly copied into your conversation. Confirm your plan’s terms.

Where you will meet it

Mostly indirectly: an “add connector” button, or a setting in Claude Code. Our class shows how tools like Claude Code connect Gmail, Drive and Slack so output lands where your team already works.

Sources

• Model Context Protocol, Introduction: “MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems”
• Model Context Protocol, Introduction: “MCP provides a standardized way to connect AI applications to external systems”
• Model Context Protocol, Architecture overview: “The AI application that coordinates and manages one or multiple MCP clients”
• Model Context Protocol, Architecture overview: “A component that maintains a connection to an MCP server”
• Model Context Protocol, Architecture overview: “A program that provides context to MCP clients”
• Model Context Protocol, Architecture overview: “Local MCP servers that use the STDIO transport typically serve a single MCP client”
• Model Context Protocol, Introduction: “Agents can access your Google Calendar and Notion”
• Model Context Protocol, Introduction: “Claude Code can generate an entire web app using a Figma design”
• Model Context Protocol, Introduction: “Enterprise chatbots can connect to multiple databases across an organization”
• Anthropic, Claude Code overview: “Claude Code can read your design docs in Google Drive, update tickets in Jira, pull data from Slack”
• Anthropic, Introducing the Model Context Protocol: “we’re sharing pre-built MCP servers for popular enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer”
• arXiv, Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions: “comprising four phases (creation, deployment, operation, and maintenance)”
• arXiv, Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions: “encompassing 16 distinct threat scenarios”
• OWASP, LLM01:2025 Prompt Injection: “These inputs can affect the model even if they are imperceptible to humans”
• Anthropic Privacy Center, Is my data used for model training? (consumer): “It does not include raw content from connectors”