AI coding agents have a context problem. They know Laravel in general. They do not know your Laravel — your schema, your registered routes, your installed package versions, your last three exceptions. So they hallucinate column names, reach for deprecated methods, and generate code that looks reasonable until it hits your actual database. The review overhead compounds. The architectural drift accumulates. You end up spending more time correcting your agent than you would have spent writing the code yourself.
The Laravel Boost MCP server is the official answer to this. Released by the Laravel team and now at v2.2.2, it is not a scaffolding tool and not a code generator. It is a Model Context Protocol server — a standardised bridge between your AI agent and your live application state. When Boost is active, your agent stops guessing and starts asking. That is a meaningful shift in the quality of what gets produced.
What “MCP” Means in Practice
MCP is an open protocol that defines how AI agents communicate with external tools and data sources. Think of it as a typed, structured API your agent can call during a task. Instead of inferring your schema from migration files it may or may not have read, an agent with Boost active can call database_schema and receive a precise, live response. Instead of assuming which routes exist, it calls list_routes. The difference between inference and interrogation is the difference between code that almost works and code that reflects the actual state of your codebase.
Boost exposes more than fifteen tools over this protocol. The ones you will reach for most often in day-to-day development:
application_info— Laravel version, PHP version, installed packages with their versionsdatabase_schema— live table structure, column types, indexes, foreign keys via Eloquent’s underlying connectiondatabase_query— read-only queries executed against your actual database, not seeded fixtureslist_routes— full route registry with middleware chains and controller bindingslast_errors— recent exceptions pulled directly from your application logtinker— PHP execution within your application’s Service Container contextsearch_docs— semantic search across versioned Laravel documentation, matched to your installed versionsread_configuration— inspect resolved config values, including environment-specific overridesbrowser_logs— capture JavaScript console output for frontend debugging loops
The search_docs tool deserves specific attention. When your agent needs to use a Cashier or Sanctum method, it does not rely on training data that may reflect a different version. It queries the documentation index for the version you actually have installed. That eliminates an entire class of subtle API mismatch bugs.
Installation
composer require laravel/boost --dev
Dev dependency only. Nothing here touches production. Once the package is installed, run the interactive setup:
php artisan boost:install
```
This is not a one-size-fits-all command. It detects your environment and walks you through selecting which IDE or agent you are integrating with: Cursor, VS Code with Copilot, Claude Code, PhpStorm with Copilot, Augment Code, Trae, or Gemini CLI. Based on your selection, Boost writes the MCP configuration to the correct path for that tool. It also publishes your AI guidelines and skills locally and registers the server. You do not write any configuration by hand.
> **[Production Pitfall]** `boost:install` writes MCP configuration to IDE-specific local paths, not into your version-controlled project directory. This means every developer on your team must run it individually. Do not assume that one person running the installer covers your whole team — it does not. Add this step explicitly to your project's onboarding documentation, and confirm it during code review if an agent-assisted PR comes in producing structurally wrong code.
---
## AI Guidelines and Skills
Beyond the MCP server itself, Boost publishes two additional layers of context.
**AI Guidelines** are markdown files that load upfront into your agent's context window before it begins any task. Boost ships guidelines for the Laravel core and a growing list of first-party packages — Livewire, Inertia, Cashier, Scout, Sanctum, Horizon, and others. Each guideline set is versioned, so the conventions your agent receives for Livewire 3 differ from those for Livewire 4. This matters more than it sounds. Subtle API changes between major package versions are exactly where agents go wrong most often, and getting that right at the context layer prevents errors that static analysis will not catch.
**Skills** are more focused: reusable, agent-readable guides for specific, repeated tasks — writing Eloquent scopes, structuring queue jobs, setting up Pest feature tests correctly. They load selectively and are designed to be high signal with low noise.
Both live locally in your project after installation. Both survive `boost:update` as long as you have not modified them — Boost will not overwrite your changes.
---
## Custom Guidelines for Your Project's Conventions
Boost's built-in guidelines cover the Laravel ecosystem. They do not cover your team's conventions. That gap is yours to fill, and it is worth filling deliberately.
Drop custom markdown files into `.ai/guidelines/` at the root of your project:
```
.ai/
guidelines/
architecture.md
naming-conventions.md
testing-standards.md
A practical architecture.md might read:
# Architecture Guidelines - Business logic lives in App\Actions — never in controllers or models. - All external API calls go through dedicated service classes in App\Services. - Controllers validate, dispatch, and respond. Nothing more. - Use typed Data Transfer Objects for inter-service data. No raw array shapes. - All queue jobs must be idempotent. Always guard against duplicate processing. - Redis is used for caching only. Do not use it as a primary data store.
The agent loads this before it writes anything. It is guidance, not enforcement — PHPStan and Pint handle enforcement — but it meaningfully reduces the rate at which agent-generated code drifts from your architectural decisions. If you are building on top of a structured service layer (if you have not set one up yet, our guide on integrating the Claude API with Laravel covers that architecture in depth), custom Boost guidelines are where you tell your agent to use it.
Wire in token tracking and rate limiting as the operational guardrail on top of that service layer.
Keeping Boost Current
php artisan boost:update
This command refreshes your published guidelines and skills against your current installed package versions. Most teams run boost:install once and forget this exists. That is a mistake. Every time you run composer update on a significant package — Livewire, Cashier, Horizon, Filament — your Boost guidelines potentially go stale. Your agent continues working from the old version context until you explicitly update.
The fix is to wire it into Composer’s post-update hook:
"scripts": {
"post-update-cmd": [
"@php artisan boost:update --no-interaction"
]
}
Idempotent, fast, and automatic. Add this once and stop thinking about it.
Team Workflow and CI Integration
Boost is a developer tooling layer, and it integrates cleanly with the broader Laravel development stack. If you are running environment-parity local setups with Herd or Sail, separated service containers for queue workers, and staging environments that mirror production — as covered in Setting Up a Laravel AI Development Stack in 2026 — Boost sits alongside all of this without conflict. The MCP server is a local Artisan process, started by your IDE when it opens the project. It does not interact with your queue workers, your Horizon dashboard, or your development server.
For CI, add the update command to your dependency installation step alongside your static analysis pipeline:
composer install --no-dev vendor/bin/phpstan analyse vendor/bin/pint --test
And in your development environment specifically:
php artisan boost:update --no-interaction
Pairing Boost’s context quality with PHPStan’s structural enforcement and Laravel Pint’s formatting consistency gives you three overlapping layers of quality control on agent-generated code. Each layer catches a different category of problem.
Common Mistakes When Adopting Boost
Skipping custom guidelines. The built-in guidelines are strong but general. If your project has architectural opinions — and it should — document them in .ai/guidelines/. An agent working without them will repeatedly make the same structural choices you would reject in code review.
Treating Boost as a substitute for review. It raises the floor on what agents produce. It does not eliminate the need for a senior developer reading what comes back. Boost improves context quality; it does not improve agent judgment.
Forgetting boost:update after package upgrades. Covered above, but worth repeating. Stale guidelines produce stale code. Automate the update.
Assuming it affects production. It does not. Boost is a dev dependency with zero production footprint. Confirm your deployment pipeline runs composer install --no-dev and move on. The service contracts and provider abstractions Boost scaffolds are covered in depth in the production-grade AI architecture guide. For the full production environment setup that Boost-generated code deploys into, see the guide on deploying Laravel to production.
The Honest Assessment
The gap between an AI agent operating blind and one with live access to your schema, routes, logs, and versioned documentation is not marginal. It is the difference between a tool that occasionally surprises you with correct output and one that is reliably useful across a full feature build. Laravel Boost closes that gap at the tooling level so you are not trying to close it through increasingly elaborate prompt engineering.
It is not magic. It is infrastructure. And like all good infrastructure, you set it up once, maintain it properly, and mostly stop thinking about it.
Official documentation:
Laravel Boost — laravel.com/docs/12.x/boost
Model Context Protocol — modelcontextprotocol.io
Senior Laravel Developer and AI Architect with 10+ years in the trenches. Dewald writes about building resilient, cost-aware AI integrations and modernizing the Laravel developer workflow for the 2026 ecosystem.