Skip to content
explainroo
Working with AI agents

What Is MCP (Model Context Protocol)?

Updated 3 min read

With explainroo How to Prompt a Coding Agent to Make an Explainer Video
On this page
  1. How an MCP connection works
  2. What MCP changes for integrations
  3. Security depends on the server and permissions

When an AI assistant needs information from a company database or needs to act in another service, it must connect to that data source or tool. MCP, or Model Context Protocol, is an open standard for those connections. It gives AI applications one way to find and use external data and tools, so developers do not need a separate integration for each assistant and service pairing.

Anthropic introduced MCP as an open-source standard in November 2024. The protocol defines how an AI application communicates with an MCP server. The server provides access to tools and information. MCP handles that exchange; the AI application still decides how to use a language model and how to present the results. Anthropic’s introduction describes the goal as connecting AI assistants to the systems where useful data and actions live.

How an MCP connection works

MCP uses three parts. The host is the AI application, such as a desktop assistant or coding environment. The host contains an MCP client, which connects to an MCP server. The server makes a service or data source available. The MCP architecture overview explains how these parts communicate.

A server can offer three kinds of capabilities. Tools perform actions, such as searching a service or updating a record. Resources provide information for the assistant to read, such as a document or database content. Prompts are reusable instructions that help structure an interaction. An assistant might use a resource to read a customer record, then call a tool to update it.

Under the hood, MCP uses JSON-RPC 2.0, a format for sending structured requests and responses. It supports local communication through standard input and output, called stdio, and remote communication through Streamable HTTP. Developers choose the transport that fits the connection: a locally running server can use stdio, while a server accessed over a network can use HTTP.

What MCP changes for integrations

Without a shared protocol, each AI application needs a custom connector for each external service. MCP gives client and server developers a common interface: an MCP-compatible client connects to compatible servers, and a server can serve compatible clients. This reduces repeated integration work, though each connection still needs configuration and the right permissions. Fast.io’s MCP overview describes this as shifting from many one-off pairings toward reusable clients and servers.

For example, a coding assistant could connect to an MCP server that exposes a code repository’s files and another that provides access to an issue tracker. The assistant could search for a bug report, inspect related code, and then draft or create an update in the tracker, depending on the tools the servers provide and the permissions the user grants.

MCP also relates to two concepts that are easy to confuse with it. Retrieval-augmented generation (RAG) retrieves relevant information from a knowledge source to help a model answer a question. MCP provides a standard way for an AI application to connect to sources and tools; a developer could use MCP to supply information for a RAG setup. Function calling lets a model request a defined action. MCP standardizes how an application connects to and discovers external capabilities, while function calling describes a model interaction. They can work together: an application can use MCP to find available tools and function calling to let the model request one. MCP and function calling covers the distinction.

Security depends on the server and permissions

An MCP connection can expose sensitive data or allow actions that change records, so operators should check what each server can access before connecting it. A server that only reads selected documents carries different risks from one that can modify customer records or run code.

The MCP security guidance warns against accepting access tokens without checking that they were intended for the receiving service. It also discusses confused-deputy attacks, in which a server is tricked into using its authority for a client without the user’s proper consent. Follow the MCP security recommendations: grant only the permissions a server needs, validate authorization tokens, and require user approval for consequential actions.

Free and open source

Let your AI agent make the video

explainroo lets a coding agent like Claude Code or Codex make narrated explainer videos and product demos. The voice, the word timing and the rendering run on your own computer, with no API key and no cost per video. Optional outside services use your own key.

Copy this into your AI agent

Make me a short explainer video about [your topic]. Use explainroo for it: run npx --yes explainroo docs and follow what it prints.

Getting started Example videos GitHub

Made with explainroo: How explainroo makes a video

More on working with ai agents

All guides on working with ai agents