WordPress

WordPress CLI permissions for AI agents

Choose what an agent may change on your WordPress site, and which checks hold whatever the settings say.

Let an AI agent work with your WordPress site through wordpress and keep the say over what goes public. The application password can do what your WordPress user can, so without your own limits an agent could do the same.

For each profile, the name you gave one site's sign-in, you decide what an agent may only read, what asks you first and what goes ahead. You can keep publishing yours with one line, and you know which checks hold whatever the settings say.

Each command has a permission key, its name such as posts.publish or api.media-delete, or a group such as api, where the longer key wins. Each key has a level: deny, readonly, ask or allow. wordpress blog config set sets them.

On this page:

  1. The levels
  2. What is set before you change anything
  3. The api group
  4. Keep publishing yours
  5. Agents and MCP
  6. Checks that hold whatever the permissions say

The levels

LevelWhat happens
denyThe command is refused, a read too.
readonlyReads work. A write is refused.
askA write asks first at a terminal. Without one it is refused unless --yes is given. Over MCP it runs.
allowThe command runs without a question.
wordpress blog config set permissions.posts.publish deny   # nothing goes public from this profile
wordpress blog config set permissions.api readonly         # every api operation reads, none writes
wordpress blog config show                                 # each key with its level and where it came from

blog is the profile, a name you chose. A refusal is permission_error (exit code 5) and names the key. It does not mean the sign-in is broken.

An agent cannot loosen a permission: a config set or config unset that would lift a limit is refused unless you run it yourself in a terminal.

What is set before you change anything

KeyBuilt-in levelChecked by
posts.createaskposts create, which makes a draft.
posts.updateaskposts update. On a published or scheduled post it checks posts.publish too.
posts.publishaskposts publish, now or with --at-time.
posts.unpublishaskposts unpublish.
posts.deleteaskposts delete, which moves a post to the trash.
posts.delete.forceaskposts delete --allow-delete, which deletes for good.
Every readallowsync, posts list, posts show and the other reads.

The api group

wordpress blog api reaches every WordPress operation, and checks the same publishing and deleting keys:

  • Publishing needs <route>.publish, such as posts.publish or pages.publish. That is a write that sets status to publish or future, or that sets date, date_gmt or password, unless the same call sets status to draft, pending or private.
  • Deleting needs <route>.delete. With force=true, which deletes for good, it needs <route>.delete.force too. Both ask by default.
  • Writes as posts commands do check their keys too. api posts-create and api pages-create check posts.create, and api posts-update and api pages-update check posts.update. So posts.create at deny stops drafts on every path.
  • Site-wide writes ask. Installing a plugin, changing the site's settings, making or changing a user (a role included), a new or changed comment, and changes to templates, template parts, widgets, sidebars, menus and navigation each ask by default: api.plugins-create, api.comments-create and so on. Other api writes, such as media, tags and categories, run without a question.
  • Making an application password is never offered: WordPress answers with the new password in clear.
  • WordPress's batch route (batch/v1) is never offered: one batch call would run other routes' writes past their keys.

wordpress blog api show <operation> lists the keys an operation checks.

Keep publishing yours

Any of these keeps your site's public face in your hands:

  • No publishing. wordpress blog config set permissions.posts.publish deny. Drafts still work. A change to a published post is refused too, as it would be public.
  • No deleting for good. wordpress blog config set permissions.posts.delete.force deny. Posts still go to the trash, from where they come back.
  • Read only. wordpress blog config set permissions.posts readonly and wordpress blog config set permissions.api readonly. Every read still works.

Agents and MCP

Over MCP nobody is at a terminal to answer, and the server shows no form, so a write at ask goes ahead. With the built-in levels an agent over MCP can publish and delete for good. The owner chose this: an agent should be able to finish the job it was given. wordpress blog runs list shows each publication for 7 days (keep_successful_runs_for_days).

If you want to publish yourself, set posts.publish to deny before you connect an app, as above. Or keep the levels and turn on approval for the wordpress_write tool in your agent's client, so the client asks you before each write.

mcp serve --permission posts.publish=deny and mcp add … --permission … make a level stricter for that one server. They never loosen the profile. See MCP.

Checks that hold whatever the permissions say

  • WCWP_READ_ONLY=1 refuses every write, for that shell or the app that sets it.
  • Your WordPress role. WordPress refuses what your user may not do, whatever the profile says. A Contributor cannot publish, for one.
  • Text from WordPress is data. Titles, text and names reach an agent marked untrusted, so an agent does not take them as instructions.

More: settings, security.