Laravel Boost: Giving Your AI Agent the Context It Actually Needs

If you have spent any time pairing with an AI coding agent on a Laravel project, you know the failure mode. You ask for a Livewire component or a Pest test, and the agent confidently produces code that looks right but uses an API that was deprecated two versions ago, or invents a method that never existed. The model was trained on a snapshot of the world that is months or years stale, and Laravel ships new features every Tuesday. Laravel Boost exists to close that gap, and after running it on real projects, it has become the first thing I install on anything I plan to build with Claude Code.

Boost is a first-party package from the Laravel team. It is not an application feature your users ever touch. It is a developer tool that gives your AI agent two things it desperately needs: accurate context about your specific project, and current knowledge about the Laravel ecosystem. It does that through an MCP server, a set of composable guidelines, on-demand agent skills, and a hosted documentation API.

Getting It Running

Installation is two commands. Add it as a dev dependency, then run the installer.

composer require laravel/boost --dev
php artisan boost:install

The installer detects which coding agents you use and generates the right configuration for each. For Claude Code that means an .mcp.json entry plus a CLAUDE.md with the relevant guidelines. Codex, Cursor, Gemini CLI, GitHub Copilot, and Junie are all supported out of the box. With Claude Code the MCP server is usually wired up automatically, but if it is not, one command fixes it:

claude mcp add -s local -t stdio laravel-boost php artisan boost:mcp

One practical tip: the generated files (.mcp.json, CLAUDE.md, AGENTS.md, and boost.json) can all go in your .gitignore. They are regenerated whenever you run boost:install or boost:update, so there is no reason to track them.

The MCP Server Is the Quiet Hero

The piece I underestimated at first is the MCP server. Boost exposes a handful of tools that let the agent inspect your actual application instead of guessing. The agent can read your PHP and Laravel versions, list installed ecosystem packages with their versions, enumerate your Eloquent models, inspect database connections, read your schema, run a read query, pull the last error from your logs, and read browser logs.

That last detail matters more than it sounds. When an agent can query your real schema, it stops hallucinating column names. When it can read the actual exception from your log file, it debugs the error you have rather than the error it imagines. I have watched Claude Code catch its own mistake by running a quick schema lookup before writing a migration, which is exactly the kind of grounding that used to require me to paste output into the chat by hand.

If you ever need to register the server manually, the configuration is trivial:

{
    "mcpServers": {
        "laravel-boost": {
            "command": "php",
            "args": ["artisan", "boost:mcp"]
        }
    }
}

Guidelines vs. Skills

Boost gives the agent ecosystem knowledge through two mechanisms, and understanding the split is worth a minute.

Guidelines are instruction files loaded upfront at the start of a session. They carry the broad, foundational conventions: how to write idiomatic Laravel, which APIs match your installed versions, when to write tests. Critically, they are version-aware. Boost ships guidelines for the framework (10.x through 13.x), Livewire (including 4.x), Inertia, Pest, PHPUnit, Pint, Tailwind, Flux UI, Pennant, and more. If you are on Livewire 4, you get Livewire 4 guidance, not a generic blend of every version the model ever saw.

Skills are the newer, more surgical approach. Where guidelines load upfront, skills load on-demand only when the agent is working on a relevant task. Boost installs them automatically based on what it finds in your composer.json. If your project has livewire/livewire, the livewire-development skill is installed. Ask Claude Code to build a Livewire component and that skill activates; move on to tests and the pest-testing skill takes over instead. The benefit is avoiding context bloat. The agent gets focused, current detail exactly when it needs it, rather than carrying every pattern for every package on every message.

You can write your own. Drop a SKILL.md into .ai/skills/{skill-name}/ for your application’s domain logic, and it gets picked up on the next boost:update. Custom guidelines work the same way through .ai/guidelines/. You can even override Boost’s built-in versions by matching the file path, which is handy when your team has opinions that differ from the defaults.

The Documentation API

The feature that earns its keep is the hosted documentation API. Boost connects your agent to over 17,000 pieces of versioned Laravel ecosystem documentation, indexed with semantic search and filtered to the major versions your project actually uses. Coverage spans the framework, Livewire, Inertia, Filament, Nova, Pest, and Tailwind.

Taylor Otwell framed the problem it solves well: Laravel releases features every Tuesday, and LLMs have no idea those features exist because they were trained on older data. Boost feeds the agent the latest docs, so it always has access to current information. The agent reaches for the Search Docs tool on its own when it is unsure, and the guidelines explicitly instruct it to do so.

Keeping It Current

Because your dependencies move, Boost can refresh its local resources to match. Run php artisan boost:update to pull the latest guidelines and skills for your installed versions. Add --discover and it will scan for newly installed packages and offer to publish their guidelines too. You can automate the whole thing by hooking it into Composer’s post-update-cmd:

{
  "scripts": {
    "post-update-cmd": [
      "@php artisan boost:update --ansi"
    ]
  }
}

Where It Fits

Boost is one of three AI packages from the Laravel team, and it is easy to confuse them. The AI SDK is for building AI features into your app. Laravel MCP is for exposing your app to external clients like ChatGPT. Boost is for the developer: it makes your coding agent write better Laravel. Interestingly, Boost is itself built on Laravel MCP, so every tool it exposes is an MCP tool under the hood.

If you write Laravel with an AI agent, install Boost first. It is a dev dependency, it never touches production, and it immediately raises the quality of what your agent produces. For a few seconds of setup, you stop fighting stale knowledge and start getting code that matches the framework you are actually running.