Back to Blog
Laravel Boost Is the Missing Piece Between AI Coding Tools and Real Laravel Apps
Laravel Boost is Laravel's first-party MCP server plus guidelines and skills for AI agents. Here's what the official docs actually ship, and where I'd still be careful.
Back
Engineering
Table of contents
Laravel Boost Is the Missing Piece Between AI Coding Tools and Real Laravel Apps
Laravel finally shipped something I've wanted since the "vibe coding" wave got serious: a first-party way for AI agents to see your Laravel app the way a senior teammate would.
It's called Laravel Boost. And if you write Laravel for a living, and you already use Cursor, Claude Code, Copilot, or another MCP-capable agent, this is worth a careful look.
Claims below are checked against the Laravel Boost docs and the Boost overview as of 18 Sep 2026. Where marketing copy and docs differ, I follow the docs.
What Boost Actually Is
Boost is not another chat wrapper.
Per the official docs, it combines:
- An MCP server so agents can inspect and interact with your app
- AI guidelines (Laravel best practices loaded up front)
- Agent skills (deeper patterns loaded on demand)
- Project rules (your team's conventions, committed under
.ai/rules) - A Documentation API with semantic search over a large Laravel-ecosystem knowledge base (docs say over 17,000 pieces of information)
Install from Composer as a dev dependency, then run the installer:
bash
composer require laravel/boost --dev
php artisan boost:installDocs also document php artisan boost:update (and optional Composer post-update-cmd / --discover) so guidelines stay aligned with the packages you actually have installed.
MCP tools (from the docs)
The Laravel 13.x Boost docs list these MCP tools:
| Tool | What the docs say it does |
|---|---|
| Application Info | PHP & Laravel versions, database engine, ecosystem packages, Eloquent models |
| Browser Logs | Browser logs and errors |
| Database Connections | Available DB connections, including default |
| Database Query | Execute a query against the database |
| Database Schema | Read the database schema |
| Get Absolute URL | Turn relative paths into absolute URLs |
| Last Error | Last error from application logs |
| Read Log Entries | Last N log entries |
| Record Rule | Persist a project rule into .ai/rules |
| Search Docs | Query Laravel's hosted docs API for your installed packages |
Note: the Boost marketing page mentions a broader set (for example route inspection, Artisan, and Tinker). I'm sticking to the documented MCP tool table above so this post doesn't overclaim.
In plain language: the agent stops inventing your folder structure and starts reading what the framework exposes.
Why This Matters More Than Another Model Release
I've written before about Cursor, Claude, Gemini, and "vibe coding." The pattern is familiar: models get better, demos look magical, then production reality shows up.
The weak point was never only "is the model smart enough?"
It was context.
Without app-aware context, AI will invent columns, miss tenant scoping, suggest patterns your codebase doesn't use, and confidently refactor the wrong layer.
Boost attacks that failure mode the way the docs describe it:
- Guidelines load Laravel ecosystem conventions up front
- Skills load deeper patterns on demand (Livewire, Inertia, Pest, Tailwind, and others listed in the docs)
- Project rules capture your conventions so new agent sessions inherit them
- Search Docs is version-aware against your installed packages
That's the difference between "AI that writes PHP" and "AI that writes your Laravel."
How I'd Use It on a Real Project
On a logistics or ERP-style Laravel codebase, I'd start small:
- Install Boost as a dev dependency (
composer require laravel/boost --dev). - Run
php artisan boost:installand enable the MCP server in Cursor or Claude Code (docs include agent-specific setup). - Ask the agent to explain an existing flow using schema + application info, not from memory.
- When the agent gets a convention wrong, record a rule (docs: ask the agent to remember it; it should use the Record Rule tool into
.ai/rules). - Keep resources current with
php artisan boost:update(or Composerpost-update-cmdas shown in the docs).
The infer-conventions skill is documented for brownfield apps: it sweeps the codebase and proposes rules from what the code already does, and it asks for approval before writing rules (unless you tell it otherwise). I'd review every proposed rule. Mixed patterns should stay mixed until the team decides, not until an agent averages them into something nobody owns.
The Parts That Make Me Careful
I'm optimistic. I'm not careless.
Database access is power. The docs explicitly include Database Schema and Database Query tools. On a machine with production credentials in .env, that is a governance problem, not a cool demo. Keep Boost on local/dev databases. Treat MCP like any other privileged local tooling.
Generated files vs. committed rules. Docs say .mcp.json, generated guideline files, and boost.json can be gitignored because they regenerate on boost:install / boost:update. Project rules under .ai/rules should be committed so the team shares them. That split matters: regenerate tooling config, version the decisions.
It won't save weak fundamentals. Boost makes agents more accurate about Laravel. It does not replace knowing why an N+1 query hurts, when a queue job should be idempotent, or how authorization should be structured. Same rule I keep repeating: AI without fundamentals just helps you ship debt faster.
Don't outsource architecture. Use Boost to accelerate boilerplate, tests, refactors, and doc lookups. Keep the hard trade-offs, boundaries, consistency, operational risk, in human hands.
Boost vs. "Just Paste More Context"
You can already dump schema.sql into chat. I've done it. It works until the dump is stale, the conversation is long, and the agent forgets half of it.
Boost is different because the tools are live and wired into the agent's loop, docs search is package-aware, rules persist across sessions and teammates, and skills load only when relevant (docs: reduces context bloat vs always-on guidelines).
That's the grown-up version of stuffing a README into every prompt.
Who Should Install It This Week
Install it if:
- You already use an MCP-capable coding agent (docs call out Cursor, Claude Code, Codex, Gemini CLI, Copilot, Junie, and others)
- Your day job is Laravel (especially mid/large apps with real conventions)
- You're tired of correcting invented models and routes
Wait (or sandbox carefully) if:
- Your only Laravel machine points at production data
- Your team has no appetite to write down conventions yet
- You're still learning Laravel itself, fundamentals first
Closing Thoughts
Laravel Boost feels like Laravel doing what Laravel does best: removing friction without pretending the hard parts disappeared.
If AI coding tools made us architects instead of typists, Boost is one of the first official tools that gives those architects a map of the actual building, and the map is written down in the docs, not only in a launch tweet.
I'll be trying it on a real project soon, not for a demo, but for the boring wins: fewer hallucinated columns, better refactors, and rules the next agent won't forget.
If you've already wired Boost into Cursor or Claude Code, tell me what broke first. Those war stories are usually more useful than the launch posts.
Sources (checked 2026-09-18)
- Laravel Boost documentation (13.x): MCP tools table, install/
boost:update, guidelines vs skills, project rules, Documentation API ("over 17,000" items) - Laravel Boost overview: product summary; broader tool marketing (not used as the MCP tool list above)
- laravel/boost on GitHub: package repository
Comments
No comments yet
Loading comments...