Laravel and AI: the official tools changing how we build (and what we build)
A practical look at Laravel Boost, the Laravel AI SDK and Laravel MCP, the first-party tools bringing AI into the Laravel ecosystem.
AI has moved from a novelty to a normal part of how web applications are built and what they can do. What’s notable in the Laravel world is that the framework team hasn’t left this to third parties. Over the past year, Laravel has shipped its own first-party AI tooling, and it now covers both sides of the equation.
On one side, there’s tooling that makes AI coding assistants genuinely good at writing Laravel code. On the other, there’s tooling that lets us build AI features directly into the applications we deliver for clients. At Enovate we’ve been working with both, and in this post we’ll walk through the key pieces and what they mean in practice.
Laravel Boost: making AI assistants write proper Laravel
AI coding assistants such as Claude Code, Cursor and GitHub Copilot are impressive, but out of the box they’re generalists. They’ll happily suggest patterns from an older Laravel version, invent a package method that doesn’t exist, or ignore the conventions your codebase already follows.
Laravel Boost tackles this head on. It’s a first-party, development-only package that gives your AI agent the context it needs to behave like an experienced Laravel developer on your project. Installation is two commands:
composer require laravel/boost --devphp artisan boost:installBoost brings four things to the table:
- An MCP server. Boost exposes tools over the Model Context Protocol that let an agent look inside your running app: reading installed package versions and Eloquent models, inspecting the database schema, running queries, and reading the latest log entries, errors and browser console output. Instead of guessing, the agent can check.
- Version-aware guidelines. Boost generates instruction files (such as
CLAUDE.mdorAGENTS.md) tailored to the exact versions of Laravel, Livewire, Inertia, Pest, Tailwind and other packages you have installed. Ask for a feature on Laravel 12 and you get Laravel 12 code. - Agent skills. Detailed, task-specific knowledge (for example, Livewire components or Pest testing) that the agent loads only when relevant, keeping its context focused.
- A documentation API. A semantic search over more than 17,000 pieces of Laravel ecosystem documentation, filtered to your installed versions.
The newest addition is arguably the most useful for an agency: project rules. These are Markdown files in a .ai/rules directory, committed to source control, that capture the decisions and conventions specific to a codebase. You can simply tell your agent “remember that money is always stored as integer cents” and Boost records it as a rule that every future agent session, and every developer on the team, inherits. There’s even an infer-conventions skill that scans an existing application and proposes rules based on what the code already does.
For a team maintaining many client applications, each with its own history and quirks, that’s a significant step. The tribal knowledge that used to live in one developer’s head now lives in the repository.
The Laravel AI SDK: AI features that feel like Laravel
Where Boost helps us build applications, the Laravel AI SDK helps us put AI inside them. Laravel announced the SDK in February 2026, it became a headline feature of Laravel 13, and it has since reached version 1.0.
The core idea is one consistent, Laravel-native API across providers. OpenAI, Anthropic, Gemini, Mistral, Bedrock, Azure, Ollama and others all sit behind the same interface, so switching models, or falling back to a second provider when the first is rate-limited or down, is a configuration change rather than a rewrite.
The central building block is the agent: a dedicated PHP class that holds its instructions, tools and output schema. Generate one with Artisan and prompt it anywhere in your app:
php artisan make:agent SupportAssistant
$response = SupportAssistant::make(user: $user) ->prompt('Where is my order?');Around that sits most of what a real feature needs:
- Tools that let an agent call your own code, such as looking up an order or checking stock.
- Structured output, so an agent returns data your application can trust, not free text you have to parse.
- Conversation memory stored in your database, plus streaming responses and queued prompts for long-running work.
- Human approval for sensitive actions: an agent can pause and wait for a person to confirm before, say, issuing a refund.
- Images, audio, transcription and embeddings, including native vector search in Eloquent on PostgreSQL (pgvector) or MariaDB, which makes retrieval-augmented generation (RAG) over a client’s own content straightforward.
- Built-in fakes for testing, so AI features can be covered by the same automated test suites as the rest of the app.
That last point matters more than it sounds. AI features are only as reliable as the engineering around them, and being able to test, queue, log and swap providers using familiar Laravel patterns is what turns a demo into something you can confidently put in front of a client’s customers.
Laravel MCP: letting AI assistants talk to your application
The third first-party piece is Laravel MCP, and it’s the one we think is most underrated. The Model Context Protocol is the open standard that lets AI assistants like Claude and ChatGPT connect to external systems. Laravel MCP lets any Laravel application become one of those systems.
In practice, you define an MCP server with tools, resources and prompts, register it like a route, and protect it with the authentication you already use (OAuth via Passport, or Sanctum tokens):
Mcp::web('/mcp/orders', OrdersServer::class) ->middleware('auth:api');A client’s customers or staff can then ask their AI assistant to check an order status, pull a report or update a booking, and the request goes through the same validation and authorisation rules as the rest of the app. Laravel MCP also works in the other direction: its client lets your own AI SDK agents use tools exposed by third-party MCP servers. Boost itself is built on it.
The wider ecosystem
The first-party tools didn’t appear in a vacuum. The community package Prism did much of the early groundwork for a unified LLM interface in Laravel, and early versions of the AI SDK built on top of it. Prism remains a solid, lighter-weight option for simpler integrations.
It’s also worth noting how these pieces reinforce each other. Boost ships guidelines and skills for Laravel MCP, and offers guided, AI-assisted upgrade prompts for the AI SDK. Agent skills follow an open standard, so the same skill can teach both the coding agent building an app and the AI agent running inside it.
What this means for Enovate’s clients
We use Laravel Boost across our Laravel projects, and project rules in particular fit the way we work. When we look after many client applications over many years, capturing each one’s conventions in the repository means faster onboarding, more consistent code and fewer “why was it done this way?” moments.
We’re also building with the AI SDK. Insights, the analytics platform we built for DestinationCore, now has an AI assistant that lets users ask questions of their data in plain English (“which sites saw the biggest drop in engagement last month?”) and get back answers, tables and charts. Under the hood it’s a single AI SDK agent class with a set of tools:
- Tools over real data. The agent can query website analytics, rank sites, break metrics down by channel or device, look up businesses and ads, and pull in context like weather and UK school holidays to help explain movements. Each tool wraps our existing, user-scoped queries, so the assistant can only ever see what that user is allowed to see.
- Charts in the conversation. A dedicated tool lets the agent render line, bar and donut charts inline in the chat, using the actual figures it has just queried.
- Structured output for follow-ups. A second, smaller agent returns a structured list of suggested next questions, constrained to data the platform actually holds.
- Production engineering around it. Turns run as queued jobs, with automatic back-off when a provider rate-limits. Token usage and cost are tracked per turn, users have a daily spend cap, and an admin view shows every conversation along with its cost and response time. The provider and model are configuration, not code.
We always pair AI-assisted development with human review and automated tests. AI makes us faster; it doesn’t replace engineering judgement, and nothing reaches a client’s production site without a developer being accountable for it.
On the product side, the AI SDK and Laravel MCP open up features that would have been expensive bespoke work not long ago. A few examples of what’s now realistic for a typical business application:
- A support assistant that answers questions using your own documentation and order data, and hands off to a person when it should.
- Automatic triage of enquiries or tickets by urgency and department.
- Summaries of long documents, transcripts or customer feedback.
- Semantic search that understands what visitors mean, not just the words they type.
- An MCP server so your team can query and update your system from the AI assistant they already use.
Because these tools are provider-agnostic, we can choose the right model for each job on cost, quality and data-handling requirements, and change that choice later without rebuilding the feature.
Getting started
AI in Laravel is no longer a collection of experiments. With Boost, the AI SDK and Laravel MCP, the framework now has a coherent, officially supported answer for building with AI and building AI into applications.
If you have a Laravel application and you’re wondering where AI could genuinely help, whether that’s a smarter search, an internal assistant or exposing your platform to AI tools, we’d love to talk it through. Get in touch with the Enovate team.
