MCP — Connecting External Tools to AI Coding Agents

Model Context Protocol is the USB-C of AI tools. Learn the protocol, wire up a server, and understand why this standard quietly changed the ecosystem.

55 min · Reviewed 2026

The Standard Nobody Expected to Matter

Model Context Protocol (MCP) is an open standard, originally published by Anthropic in late 2024, that lets AI agents talk to external tools using a common JSON-RPC interface. By April 2026 it has hit roughly 100 million monthly downloads with 3,000+ indexed servers and is supported by Claude Code, Codex CLI, Cursor, and Windsurf.

Why this matters

Before MCP, every AI tool invented its own plugin system. If you wrote a GitHub integration for one IDE, it didn't work anywhere else. MCP is the common adapter — one server speaks to every compliant client. The USB-C analogy is apt.

The three primitives

Tools: functions the agent can call (query_db, create_issue, send_slack_message)
Resources: read-only data sources the agent can pull from (files, docs, schemas)
Prompts: reusable prompt templates the server exposes as commands

A minimal MCP server in TypeScript

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "weather",
  version: "1.0.0",
});

server.tool(
  "get_forecast",
  {
    city: z.string().describe("City name"),
  },
  async ({ city }) => {
    const res = await fetch(`https://api.weather.example/${city}`);
    const data = await res.json();
    return {
      content: [{ type: "text", text: JSON.stringify(data) }],
    };
  },
);

const transport = new StdioServerTransport();
await server.connect(transport);A real MCP server in ~20 lines. Any compliant client can now call get_forecast with a city string.

Wiring a server to a client

{
  "mcpServers": {
    "weather": {
      "command": "node",
      "args": ["/path/to/weather-server/index.js"]
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}A claude_desktop_config.json or .claude/mcp.json file wiring two servers. Same file format across Claude Code, Cursor, and Windsurf.

Servers worth knowing in 2026

Category	Example servers
Databases	Postgres, SQLite, MongoDB, Redis
Version control	GitHub, GitLab
Docs	Sentry, Linear, Notion, Confluence
Cloud	AWS, Vercel, Cloudflare, Stripe
Browsers	Playwright, Puppeteer, Chrome DevTools

Security model

Servers run locally by default — no mystery cloud calls
The client mediates every tool invocation — user can approve or deny
Auth flows through the server (API keys, OAuth) — not exposed to the model
Audit logs of tool calls are the minimum observability you should require

Protocols are patient. They don't win today; they win in five years.
— A veteran protocol designer

The big idea: MCP turned AI coding tools from siloed islands into a composable ecosystem. If you build, extend, or integrate AI agents in 2026, this is the protocol layer you cannot ignore.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-coding-mcp-external-tools-creators

What problem was MCP designed to solve in the AI tooling landscape?
1. AI coding tools were too slow without a protocol layer
2. MCP was created to replace JSON-RPC as a communication standard
3. AI models couldn't communicate with each other directly
4. Each AI tool required its own unique plugin system that couldn't work elsewhere
What communication interface does MCP use to let AI agents talk to external tools?
1. GraphQL
2. REST API
3. SOAP
4. JSON-RPC
Which company originally published the Model Context Protocol?
1. OpenAI
2. Microsoft
3. Google
4. Anthropic
What does the 'USB-C analogy' refer to in the context of MCP?
1. MCP was designed to replace older USB protocols
2. MCP can only be used with hardware that has USB-C ports
3. MCP serves as a universal adapter allowing different tools to connect to AI agents
4. MCP uses the same physical connector as USB-C devices
What are the three primitive categories defined in MCP?
1. Services, handlers, and listeners
2. Plugins, extensions, and add-ons
3. APIs, endpoints, and functions
4. Tools, resources, and prompts
In MCP terminology, what are 'Tools'?
1. Functions that an AI agent can actively invoke to perform actions
2. Authentication credentials for accessing servers
3. Read-only data sources like files and documentation
4. Reusable prompt templates exposed by the server
What type of data can MCP 'Resources' provide to an AI agent?
1. Authentication tokens for API access
2. Read-only data sources like files, documents, and schemas
3. Write access to modify external systems
4. Executable code that the agent can run
What are MCP 'Prompts' used for?
1. Providing temporary storage for agent conversations
2. Authenticating users to MCP servers
3. Encrypting communication between client and server
4. Exposing reusable prompt templates as commands on the server
What is a key security advantage of MCP servers running locally by default?
1. They eliminate mystery cloud calls that users can't see
2. They automatically encrypt all data at rest
3. They don't require any authentication
4. They can bypass firewall restrictions
Where does authentication flow in MCP's architecture?
1. Through the MCP server, not exposed to the model
2. Through a third-party identity provider
3. Through the AI model directly
4. Through the client application
Why should you audit the source code of MCP servers before installing them?
1. MCP servers are always open source by license requirement
2. Source code audit is required by the JSON-RPC standard
3. MCP servers require a paid subscription to use safely
4. Installing a server installs code that runs with your privileges
What does the lesson identify as the minimum observability requirement for MCP deployments?
1. Audit logs of tool calls
2. Video recordings of user sessions
3. Network traffic encryption logs
4. Real-time dashboards showing server health
Which permission model does MCP support for agents attempting to modify production systems?
1. MCP servers can require explicit user approval before write operations
2. Agents always have full write access by default
3. Write access is granted only through a separate authentication protocol
4. Production modifications are blocked at the network level
What architectural change did MCP bring to AI coding tools?
1. It made all AI models run locally
2. It turned siloed islands into a composable ecosystem
3. It replaced all existing IDEs with cloud-based alternatives
4. It eliminated the need for programming languages
By April 2026, approximately what was MCP's monthly download metric?
1. 100 million
2. 1 billion
3. 10 million
4. 500 million

← Back to interactive lesson

Tendril · Creators · AI-Assisted Coding

MCP — Connecting External Tools to AI Coding Agents

Model Context Protocol is the USB-C of AI tools. Learn the protocol, wire up a server, and understand why this standard quietly changed the ecosystem.

55 min · Reviewed 2026

The Standard Nobody Expected to Matter

Why this matters

The three primitives

Tools: functions the agent can call (query_db, create_issue, send_slack_message)
Resources: read-only data sources the agent can pull from (files, docs, schemas)
Prompts: reusable prompt templates the server exposes as commands

A minimal MCP server in TypeScript

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "weather",
  version: "1.0.0",
});

server.tool(
  "get_forecast",
  {
    city: z.string().describe("City name"),
  },
  async ({ city }) => {
    const res = await fetch(`https://api.weather.example/${city}`);
    const data = await res.json();
    return {
      content: [{ type: "text", text: JSON.stringify(data) }],
    };
  },
);

const transport = new StdioServerTransport();
await server.connect(transport);A real MCP server in ~20 lines. Any compliant client can now call get_forecast with a city string.

Wiring a server to a client

{
  "mcpServers": {
    "weather": {
      "command": "node",
      "args": ["/path/to/weather-server/index.js"]
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}A claude_desktop_config.json or .claude/mcp.json file wiring two servers. Same file format across Claude Code, Cursor, and Windsurf.

Servers worth knowing in 2026

Category	Example servers
Databases	Postgres, SQLite, MongoDB, Redis
Version control	GitHub, GitLab
Docs	Sentry, Linear, Notion, Confluence
Cloud	AWS, Vercel, Cloudflare, Stripe
Browsers	Playwright, Puppeteer, Chrome DevTools

Security model

Servers run locally by default — no mystery cloud calls
The client mediates every tool invocation — user can approve or deny
Auth flows through the server (API keys, OAuth) — not exposed to the model
Audit logs of tool calls are the minimum observability you should require

Protocols are patient. They don't win today; they win in five years.
— A veteran protocol designer

The big idea: MCP turned AI coding tools from siloed islands into a composable ecosystem. If you build, extend, or integrate AI agents in 2026, this is the protocol layer you cannot ignore.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-coding-mcp-external-tools-creators

What problem was MCP designed to solve in the AI tooling landscape?
1. AI coding tools were too slow without a protocol layer
2. MCP was created to replace JSON-RPC as a communication standard
3. AI models couldn't communicate with each other directly
4. Each AI tool required its own unique plugin system that couldn't work elsewhere
What communication interface does MCP use to let AI agents talk to external tools?
1. GraphQL
2. REST API
3. SOAP
4. JSON-RPC
Which company originally published the Model Context Protocol?
1. OpenAI
2. Microsoft
3. Google
4. Anthropic
What does the 'USB-C analogy' refer to in the context of MCP?
1. MCP was designed to replace older USB protocols
2. MCP can only be used with hardware that has USB-C ports
3. MCP serves as a universal adapter allowing different tools to connect to AI agents
4. MCP uses the same physical connector as USB-C devices
What are the three primitive categories defined in MCP?
1. Services, handlers, and listeners
2. Plugins, extensions, and add-ons
3. APIs, endpoints, and functions
4. Tools, resources, and prompts
In MCP terminology, what are 'Tools'?
1. Functions that an AI agent can actively invoke to perform actions
2. Authentication credentials for accessing servers
3. Read-only data sources like files and documentation
4. Reusable prompt templates exposed by the server
What type of data can MCP 'Resources' provide to an AI agent?
1. Authentication tokens for API access
2. Read-only data sources like files, documents, and schemas
3. Write access to modify external systems
4. Executable code that the agent can run
What are MCP 'Prompts' used for?
1. Providing temporary storage for agent conversations
2. Authenticating users to MCP servers
3. Encrypting communication between client and server
4. Exposing reusable prompt templates as commands on the server
What is a key security advantage of MCP servers running locally by default?
1. They eliminate mystery cloud calls that users can't see
2. They automatically encrypt all data at rest
3. They don't require any authentication
4. They can bypass firewall restrictions
Where does authentication flow in MCP's architecture?
1. Through the MCP server, not exposed to the model
2. Through a third-party identity provider
3. Through the AI model directly
4. Through the client application
Why should you audit the source code of MCP servers before installing them?
1. MCP servers are always open source by license requirement
2. Source code audit is required by the JSON-RPC standard
3. MCP servers require a paid subscription to use safely
4. Installing a server installs code that runs with your privileges
What does the lesson identify as the minimum observability requirement for MCP deployments?
1. Audit logs of tool calls
2. Video recordings of user sessions
3. Network traffic encryption logs
4. Real-time dashboards showing server health
Which permission model does MCP support for agents attempting to modify production systems?
1. MCP servers can require explicit user approval before write operations
2. Agents always have full write access by default
3. Write access is granted only through a separate authentication protocol
4. Production modifications are blocked at the network level
What architectural change did MCP bring to AI coding tools?
1. It made all AI models run locally
2. It turned siloed islands into a composable ecosystem
3. It replaced all existing IDEs with cloud-based alternatives
4. It eliminated the need for programming languages
By April 2026, approximately what was MCP's monthly download metric?
1. 100 million
2. 1 billion
3. 10 million
4. 500 million

← Back to interactive lesson