laravel-13-ai-boost-mcp

Laravel 13 AI, Boost, and MCP

Workflow

1. Check whether Laravel Boost is installed and available.
2. Use Boost/search-docs or official package docs for exact APIs when available.
3. Identify whether the task is app AI feature, AI SDK integration, MCP server/tool, or agent workflow.
4. Treat AI boundaries like external services: validate inputs, constrain outputs, fake in tests, and avoid leaking sensitive data.
5. Test deterministic behavior around prompts, tools, schema validation, and fallback paths.

Agent Compatibility (Cursor, Codex, Claude Code)

• Treat "Boost" as optional. If unavailable, rely on composer.json + official docs and avoid guessing fast-moving APIs.
• Keep MCP tool specs self-describing (name, purpose, input schema, output shape) so they work in any agent runtime.

Defaults

• Prefer official Laravel packages and patterns over hand-rolled clients.
• Use structured output schemas when downstream code depends on AI output.
• Keep prompts versioned and close to the feature.
• Use fakes/mocks for AI providers in tests.
• For MCP tools, define clear input schemas, narrow capabilities, authorization, and safe error messages.
• For embeddings/vector search, document model, dimensions, storage, indexing, and refresh strategy.
• For package-style AI/MCP integrations, prefer PSR-3 logging, PSR-18 HTTP clients, PSR-20 clocks, and PSR-11 container boundaries when interoperability is useful.

Security + Cost Controls (default stance)

• Never pass secrets/PII to providers unless strictly necessary and permitted.
• Add guardrails against prompt injection and tool overreach (narrow tools, validate inputs, validate outputs).
• Bound execution: max tokens, max tool calls, timeouts, retries, and budget controls when available.

Reference

Read reference.md for detailed patterns, checklists, and examples for this skill.