> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tydli.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Core Concepts

> Understanding the Model Context Protocol

# Core Concepts

## What is the Model Context Protocol?

The **Model Context Protocol (MCP)** is an open standard created by Anthropic that enables AI systems to securely connect with external tools, data sources, and APIs.

### Think of MCP as USB-C for AI

Just like USB-C provides a universal standard for connecting devices, MCP provides a universal standard for AI agents to connect with external tools and APIs. One protocol, any AI system, any API.

## MCP Architecture Flow

The MCP ecosystem consists of three key components working together:

```
AI Host (Claude, ChatGPT, Custom Agent)
    ↓
MCP Client (Protocol Translator)
    ↓
MCP Server (Your API via Tydli)
```

### MCP Host (AI Application)

The AI application that wants to use external tools - like Claude Desktop, ChatGPT, or a custom AI agent you've built.

**Analogy:** Think of the Host as the "boss" who needs to delegate specific tasks to specialized workers. The boss knows what needs to be done but doesn't do the actual work.

**Examples:**

* Claude Desktop application
* ChatGPT with custom plugins
* Custom AI assistants built with LangChain
* Continue.dev for VS Code

### MCP Client (Protocol Bridge)

The connector that speaks both languages - it translates between what the AI Host wants and what the MCP Server provides.

**Analogy:** Like a translator at a business meeting. The Host speaks "AI language" and the Server speaks "API language." The Client makes sure both sides understand each other perfectly.

**What it does:**

* Discovers available tools from MCP Servers
* Translates AI requests into proper API calls
* Handles authentication and security
* Manages errors and retries

### MCP Server (Your API)

Your actual API wrapped in MCP protocol. This is what Tydli creates for you automatically from your OpenAPI spec.

**Analogy:** The specialized worker who does the actual job. When asked to "get weather data," it knows exactly how to fetch it from the weather API and return it in a format the AI can understand.

**Capabilities:**

* **Tools:** Functions that perform actions (e.g., "send\_email", "get\_user")
* **Resources:** Data sources to read from (e.g., "user\_database", "file\_storage")
* **Prompts:** Reusable templates (e.g., "professional\_email\_template")

## Why MCP Changes Everything

### The Problem Before MCP

* ✗ Each AI system had its own custom integration format
* ✗ Building plugins required weeks of development time
* ✗ APIs built for one AI didn't work with others
* ✗ Security and authentication were inconsistent
* ✗ Maintaining multiple integrations was a nightmare

### The Solution With MCP

* ✓ One standard protocol works across all AI systems
* ✓ Deploy in seconds with tools like Tydli
* ✓ Write once, use with Claude, ChatGPT, and more
* ✓ Built-in security and authentication patterns
* ✓ Automatic updates without breaking changes

## How Tydli Fits In

Tydli automatically generates MCP Servers from your existing OpenAPI specifications. This means:

1. **No code required** - Your OpenAPI spec is all you need
2. **Instant deployment** - Get an MCP server URL in seconds
3. **Production-ready** - Built-in authentication, rate limiting, and error handling
4. **MCP compliant** - Follows the official Model Context Protocol 2025-03-26 specification

## Next Steps

* Set up your first MCP server with our [Quickstart Guide](/getting-started/quickstart)
* Learn how to [connect with Claude Desktop](/guides/oauth-integration)
* Explore [authentication options](/guides/authentication-methods)
* Read about [best practices](/guides/best-practices) for production deployments