Security Checklist

When to Use

Run through this before exposing Tool API tools to any non-human caller. Each item links to the page that owns the rule.

Version: applies to drupal/tool 1.0.0-beta11 (beta; no security advisory coverage), tool_belt 1.0.0-alpha6, mcp_server_tool_bridge 1.0.0-beta3.

Checklist

• [ ] Every tool declares a permission for account-level rules and uses checkAccess() for object-level rules (OWASP API1, broken object level authorization) → Access Control
• [ ] checkAccess() passes the given $account to every entity and field access call → Access Control
• [ ] Invokers you write call checkPermission() before setting inputs, and validateInputs() before access() → Calling a Tool from PHP
• [ ] MCP-exposed tools keep refiners and input transforms free of side effects, because the bridge sets inputs before any denial → Calling a Tool over MCP
• [ ] Failure messages and failure context_values carry no internals → doExecute and ExecutableResult
• [ ] No undeclared result keys carry entities or secrets → Output Definitions
• [ ] Write tools validate the entity before saving; tool_belt:entity_save does not → Tool Belt Catalog
• [ ] operation and destructive are honest, and agent tool lists exclude destructive tools → Operation and Destructive, Calling a Tool from the AI Module
• [ ] Prerequisites that must block are re-checked in doExecute() → checkRequirements
• [ ] Access rules are tested with a role-scoped --uid, not uid 1 → Calling a Tool from Drush
• [ ] administer tool stays on trusted roles → Installation and Submodules

Common Mistakes

• Trusting operation: Read to mean safe → nothing enforces it; review the code

See Also

• Access Control
• OWASP API Security Top 10: https://owasp.org/API-Security/