Learn › AI Agents: Core Capabilities › MCP and Agent Protocols

MCP and Agent Protocols

Intermediate 🕐 12 min Lesson 7 of 7
What you'll learn
  • Explain what MCP is and what problem it solves
  • Describe the three-layer MCP architecture (host, client, server)
  • Name the three MCP primitives and what each one does

The Integration Problem

Before MCP, every AI tool integration was custom. If you wanted your agent to read Notion pages, query a database, search Slack, and call a CRM API, you had to write four separate integrations — each with its own authentication, data format, and error handling. Add a new agent framework, and you might need to rewrite all four.

MCP is an attempt to solve this with a standard protocol: build a tool once, and it works with any agent that speaks MCP.

What MCP Is

Model Context Protocol (MCP) is an open standard introduced by Anthropic in late 2024 and donated to the Linux Foundation in December 2025. It defines a common way for AI applications (clients) to connect to external tools and data (servers).

The key idea: instead of each AI app and each tool negotiating their own interface, everyone uses the same protocol. Tool developers write one MCP server. Agent developers write one MCP client. They work together by default.

The Client-Server Architecture

MCP uses a standard client-server model:

  • MCP Host — the application the user interacts with (Claude Desktop, Cursor, your custom agent)
  • MCP Client — the protocol client running inside the host, managing connections to servers
  • MCP Server — a lightweight process that exposes one or more capabilities

A host can connect to many servers simultaneously. The agent discovers what each server offers and can call any capability at runtime.

The Three Primitives

MCP servers expose capabilities through three structured types:

  • Tools — callable functions with side effects (search the web, write a file, send a message). The agent calls these when it needs to take an action.
  • Resources — read-only data sources with URI addresses (a file, a database row, a web page). The agent reads these when it needs context.
  • Prompts — reusable templates for common tasks. These can be surfaced to users as workflows or invoked by the agent directly.

Transport and Security

MCP servers communicate over two transports: stdio (local process, lowest latency) and HTTP (remote server, enables cloud-hosted tools). Authentication and authorization are handled by the server — each server decides which clients can access which capabilities.

Why It Matters

Standardization compounds. When a tool developer publishes an MCP server for their product, every agent that speaks MCP can use it immediately. As of 2026, major frameworks — LangChain, AutoGen, CrewAI — have added MCP support, and many tool providers have published MCP servers. The ecosystem is early but growing fast.

For builders: wrapping your internal tools as MCP servers makes them available to any future agent you build, without rewriting the integration each time. It's the kind of infrastructure investment that pays off slowly and then all at once.

Key takeaways
  • MCP solves the N×M integration problem: build a tool once as an MCP server and it works with any MCP-compatible agent
  • The three MCP primitives are: tools (callable functions with side effects), resources (read-only data with URIs), and prompts (reusable templates)
  • Anthropic introduced MCP in late 2024 and donated it to the Linux Foundation in December 2025 — no single company controls the protocol
  • MCP runs over two transports: stdio for local servers and HTTP for remote; authentication is handled server-side
Learn more every week Get 5 hand-picked prompts + new AI tools every Monday. Free.