Model Context Protocol

The Model Context Protocol (MCP) is the emerging standard for bridging the gap between AI assistants and your custom applications. Developed by Anthropic but adopted by OpenAI and others, it provides a unified language for Claude or ChatGPT to understand your database and logic. Laravel MCP streamlines this by providing a first-party package to turn your application into an AI-ready server. Prerequisites Before diving in, you should have a solid grasp of Laravel, particularly how to handle authentication with Laravel Sanctum. You will also need an MCP-compatible client like Cursor or VS Code to test your server. Key Libraries & Tools * Laravel MCP: The core package for building MCP servers. * Laravel Sanctum: Handles the secure token-based authentication required by AI agents. * Locket MCP: A demo application showcasing real-world MCP integration. The Three Pillars: Prompts, Resources, and Tools To build a functional MCP server, you must implement three core components using `php artisan make` commands. Prompts as System Instructions Prompts act as specialized system instructions. They tell the AI exactly how to interpret user intent within your app's context. For instance, a `SummarizeLink` prompt might define how the AI should analyze a URL before storing it. Resources for Data Retrieval Resources represent the read-only state of your application. They are perfect for providing the AI with data from your database, such as the "last added link." ```php public function handle() { return Auth::user()->links()->latest()->first(); } ``` Tools for Actionable Logic Tools are the "bread and butter" of MCP. Unlike resources, tools use an input schema to perform actions or complex queries. If a user asks to "Add this link," the AI uses a tool with a defined schema to pass the URL to your `handle` method. Syntax Notes & Best Practices Every MCP component requires a `handle` method and a detailed description. The AI relies on these descriptions to decide which tool to call. Use the `inputSchema` in your tools to enforce constraints, such as limiting the number of records returned to prevent token overflow. Tips & Gotchas Authentication is often the trickiest part. Laravel MCP uses Laravel Sanctum out of the box, allowing for a seamless "Connect" flow within clients like Cursor. Always test your tools locally before deploying to ensure the AI understands your schema descriptions.