Skip to main content

Security & Permissions

WP-MCP is built security-first. Every AI request goes through a multi-step pipeline before anything happens on your site.

The security pipeline​

Every single tool call goes through these checks in order:

  1. Plugin enabled check — is WP-MCP turned on?
  2. Authentication — valid Application Password or Bearer Token?
  3. Rate limiting — under the per-minute request limit?
  4. Tool enabled — not disabled by the admin?
  5. Permission profile — does this user/client have access to this tool level?
  6. WordPress capability — does the user have the required WordPress permission?
  7. DANGER confirmation — if the tool is DANGER level, was confirm: true provided?
  8. Parameter validation — are all required fields present and valid?
  9. Execute — run the tool
  10. Audit log — record the action

Authentication methods​

Standard WordPress authentication. Create one password per AI client so you can revoke access individually without affecting other connections.

Generate from the WP-MCP Dashboard — one click, no profile page needed.

Cowork Bearer Token​

An alternative token-based authentication for AI clients that can't use Application Passwords. Generate from WP-MCP → Security.

Permission profiles​

Assign a different access level per user and per AI client:

ProfileREAD toolsWRITE toolsDANGER tools
Full access✅✅✅
Write access✅✅❌
Read only✅❌❌
Blocked❌❌❌

Configure from WP-MCP → Security → Permission Profiles.

DANGER tool confirmation​

Tools marked 🔴 DANGER (delete, bulk changes, critical settings) require confirm: true in the request. This is enforced at the plugin level — Claude or Codex will always ask you to confirm before proceeding.

You can disable this in WP-MCP → Settings, but it's strongly not recommended.

Rate limiting​

Default: 60 requests per minute per user. Adjust from WP-MCP → Settings.

Audit logging​

Every action is logged — tool name, user, timestamp, parameters, and result. View the full log at WP-MCP → Audit Log.

Best practices​

  • Create one Application Password per AI client — name them "WP-MCP Claude", "WP-MCP Codex"
  • Assign Write access (not Full access) to clients that don't need to delete anything
  • Keep DANGER confirmation enabled — the extra step takes 2 seconds and prevents accidents
  • Review the Audit Log periodically to see what your AI assistant has been doing
  • Revoke Application Passwords immediately if you suspect unauthorized access
Last updated on