- What is mnapoli/backoffice-mcp and what does it do?
- It is a Model Context Protocol server for Backoffice that gives AI agents access to your app's models for admin and support tasks. In Laravel/PHP apps it exposes standardized MCP tools and resources so assistants can interact with back-office tooling. Treat it as an integration layer between your application and an MCP-capable AI client.
- How do I install mnapoli/backoffice-mcp in a Laravel application?
- Install it with Composer, then confirm whether the package registers a service provider and publishes configuration using standard Laravel conventions. The assessment notes these items should be verified because no prior detailed evaluation exists. After installation, define a minimal server description and test against your actual MCP client in staging.
- Which Laravel and PHP versions does mnapoli/backoffice-mcp support?
- The available context does not list exact Laravel or PHP version constraints, so check the package's composer.json and README before installing. Confirm compatibility with your framework version, PHP runtime, and MCP client. Because 0.2.0 is pre-1.0, pin the version you validate rather than assuming semver stability.
- How do I set the server $description that the AI agent sees?
- In 0.2.0 you define a server-level $description on the server class, and the package passes it to the agent before any generic instructions. Keep it concise and accurate about your domain models, permissions, and data semantics. Review it whenever models or business rules change because it becomes agent-facing documentation.
- Can I disable or override the generic instructions in mnapoli/backoffice-mcp?
- The assessment explicitly flags this as an open question: confirm what the generic instructions contain and whether they can be disabled or overridden. The generic guidance reportedly directs the agent to inspect application code, which may be unsuitable in production or least-privilege environments. Test instruction precedence and consider wrapping or replacing the behavior if the package allows it.
- Is mnapoli/backoffice-mcp safe for production admin and support use?
- The main risks are prompt and context risks rather than PHP runtime risks. A stale or misleading $description can cause the agent to make incorrect assumptions, and broad code-inspection instructions may expose secrets or architecture details if the agent has code access. Run a security review for prompt injection, data exposure, and least privilege before production use.
- How should I test mnapoli/backoffice-mcp before deploying to production?
- Test with the actual MCP agent client, not only unit tests, because behavior depends on prompt ordering and context. Use representative app data in staging, check how the server $description affects responses, and verify instruction precedence. Add regression cases for prompt-injection attempts and sensitive-data exposure.
- What changed in mnapoli/backoffice-mcp 0.2.0 and is it a breaking change?
- 0.2.0 adds a server-level $description that is sent to the agent before generic instructions. The change appears additive, but 0.2.0 is pre-1.0, so minor releases may still contain breaking changes. Review the changelog and migration notes from 0.1.x, then pin the version you validate.
- Does mnapoli/backoffice-mcp support per-tenant or localized server descriptions?
- The assessment lists per-tenant overrides, localization, sanitization, and length limits as questions to verify rather than confirmed features. Check the package source and documentation for how $description is stored and passed. If multi-tenancy matters, test whether the description can vary by environment or tenant without leaking architecture details.
- What are alternatives to mnapoli/backoffice-mcp for AI admin access in Laravel?
- The provided context does not compare alternatives, so evaluate it against other Model Context Protocol servers and Laravel admin tooling based on your needs. Look for packages that follow Laravel conventions, document prompt safety, and support the MCP client you use. Because this is a greenfield evaluation, run a small proof of concept before committing.