Free on WordPress.org, built on the Abilities API

Let ChatGPT and Claude work in WordPress, on a leash you hold.

Agent Abilities for MCP is a free plugin that lets your AI assistant draft posts, tidy media, update WooCommerce and fill in SEO fields, signed in as a WordPress user you pick rather than an admin key. Every ability starts off, and every call to its tools is checked and logged, which makes your site a governed WordPress MCP server.

Works with ChatGPT (as a custom MCP server on paid plans), Claude.ai, Claude Desktop, Manus, Claude Code and 4 more. How to connect

  • Free on WordPress.org
  • No AI provider contacted
  • No telemetry
  • Off by default
  • Every call to its tools audited
  • Undo for 25 of its tools
Your AI client, signed in as a Contributor Illustrative

An illustrative run built from the plugin’s real tool names and Activity Log columns. The request, times and ids are made up. The agent only ever hears “Permission denied”; the reason stays in your log.

1,000+ active installs, and still early.

Every number here comes from WordPress.org, checked 17 November 2026. The quotes are copied word for word from public reviews.

  • 1,000+ active installs on WordPress.org
  • 4.8 of 5 from 5 ratings
  • 3 of 3 support threads resolved, last two months
  • Listed in the official MCP Registry
  • Stumbled onto this gem thanks to ChatGPT. It made managing multiple websites easier! A time saver.

    ★★★★★ danielkam1 on WordPress.org, 9 October 2026
  • I’ve been using this plugin across several WordPress sites, and it’s worked well for me. Setup was straightforward, and handling WooCommerce products and orders with it has saved me a lot of time. Happy to give it five stars.

    ★★★★★ davidbaracsy1992 on WordPress.org, 7 October 2026
  • I was able to connect MCP server with claude via WordPress Application Password but was unable to connect it with chatgpt as it allows only OAuth authentication so I was trying to add OAuth authentication in my custom abilities plugin but I found this plugin which saved me with this extra work.

    ★★★★★ Muhammad Farooq Abdullah on WordPress.org, 15 July 2026

Read all 5 reviews on WordPress.org

Connect ChatGPT, Claude or your code editor in three steps.

The plugin installs with every ability off. When you are ready, this is the whole setup, and from then on the agent works with that WordPress user’s permissions.

  1. Turn on what you want

    Open the Agent Abilities for MCP screen and enable only the abilities you want the agent to have. Everything starts off.

  2. Copy your endpoint

    On the Connection tab, copy your site’s MCP endpoint.

  3. Add it to your AI client

    In ChatGPT, Claude.ai, Claude Desktop or Manus, add it as a custom connector and approve once over OAuth. Code editors and command-line clients take the same endpoint, over OAuth or an Application Password.

ChatGPT connects as a custom MCP server on its paid plans, with write tools on Business and Enterprise/Edu. The Gemini app has no known way to add a custom MCP server.

OAuth or an Application Password? How connecting works. Requires WordPress 6.9+ and PHP 7.4+. Free on WordPress.org, no paid tier, no API key to buy, no usage limits.

Undo what an agent changed.

For edits made with 25 of this plugin’s tools, the Changes tab keeps up to 30 days and puts a change back if the item hasn’t been edited since. Every write tool says in its response whether its change can be undone.

A tab of every change

Each covered edit gets a row with the tool, the item and the field names it touched. Revert puts it back, and Revert all puts back a site-wide find and replace.

Agents undo their own work

Two optional tools, both off by default, let an agent undo its own changes from the last 24 hours, up to 30 undo calls an hour per connection.

It says what it can’t undo

Permanent deletes, orders, refunds, user accounts, file uploads and other plugins’ tools are not covered, and a revert that would overwrite a newer edit is refused.

The Changes tab in wp-admin: a table of changes with When, Tool, Item, Fields, By, Status and Action columns, most rows showing a Can revert status and a Revert button, and one replace-sitewide row with a Revert all button
The Changes tab on the demo site, taken while testing 1.11.2. A real screenshot of the plugin admin.

The rules every connection follows.

These 6 guarantees hold on every connection and every call to its tools, not just the first one.

Least privilege by design

The agent connects as a real WordPress user, never an admin-equivalent key, and gets only that account’s capabilities.

Off by default

Nothing this plugin serves is exposed until you enable it, and an update never switches on a tool it serves.

Two-layer capability gating

A connection only sees the plugin’s own tools that its user can call, and every call re-checks that capability before it runs.

Honest audit log

Every call to the tools it serves is logged in your own database, denials included with the reason why, identifiers only and never free-text content. Search it, filter it and export it.

Bounded by construction

No arbitrary option or meta access, and no code execution. Permanent deletes ship off, and the last administrator can never be removed.

No content leaves your site

The plugin contacts no AI provider and sends no telemetry. Your AI client connects in, and no site content goes out on its own.

What happens on every call

Discovery and execution are gated separately, so an agent never gets more reach than the user you bound it to.

  1. Connect as a scoped user

    The agent signs in as a real WordPress account you pick, never an admin key.

  2. See only what that user can call

    The plugin’s own tools are listed only to users allowed to run them. Bridged tools whose own check needs an item id list for anyone who can edit posts, and the source plugin decides each call.

  3. Re-check on every call

    The plugin checks that user’s capability again before any ability runs.

  4. Record it, denials included

    Every call to its tools lands in a searchable log on your own site, and a refusal records why.

  5. Stay bounded by construction

    No arbitrary option access and no code execution. Permanent deletes ship off.

Optional guardrails, all off until you set them:Per-minute rate limit offIP allowlist offForce-to-draft mode offTitle-length cap offBlock-markup validation (warn or strict) off

216 governed abilities, and growing.

Curated WordPress actions, exposed to the agent as MCP tools: drafts and edits, terms and media, menus and blocks, SEO fields and store data. Each one stays off until you switch it on, so nothing runs that you did not choose.

Bridge other plugins’ abilities

Your other active plugins can register their own WordPress abilities. This plugin brings them in as governed MCP tools, each one off until you switch it on. They come in through the WordPress Abilities API, and sit on top of the 216 in the catalog, off until you choose it. Since 1.11.0 the tab for other plugins’ tools (named Plugins’ own tools since 1.12.0) says which plugin provides each group of tools, labels each tool Read, Write or Destructive, and lists tools a plugin adds later for your review.

See how bridging works

Browse the full catalogSee what people actually run

Write the steps once. Every agent follows them.

Playbooks are step-by-step instructions you save in WordPress, such as how to check a draft before it goes live. Claude, ChatGPT and your other clients load them from your site, so a job gets done the same way whoever asks.

  • Works from Claude and ChatGPTThe agent lists your Active playbooks, loads one and works through it, using only the tools you switched on.
  • Fine on a limited connectionSince 1.10.2, Content editor and Shop manager connections can read playbooks too, as well as Read only and Custom.
  • No extra accessA playbook turns nothing on, and agents can read playbooks but never edit one.
  • 5 starters, 20 more to copyThe starters ship with the plugin. The library adds more, grouped by job: content, SEO, WooCommerce, events and others.

Starter playbook

Publish checklist

Writes after you say yes

Check one draft before it goes live and list what is missing or worth fixing. Changes nothing unless you say yes. Never publishes.

Get postsGet postGet termsGet mediaSet featured imageAdd post terms

Use this when the user asks you to check a draft before it is published.

1. Find the draft. If the user named it or gave a link or ID, read it with the Get post tool. If not, list recent drafts with the Get posts tool (status draft) and ask which one to check. Do not guess.

2. Title. Say so if it is longer than about 60 characters, is vague, or does not match what the post is about.

It all runs from one screen in your admin.

No external dashboard, no account with us. The plugin lives in wp-admin: your endpoint, the abilities you enabled, the agent users, and the audit log, all in your own database.

  • Your MCP endpoint
  • The abilities you enabled
  • The agent users
  • The audit log
The Agent Abilities for MCP dashboard in wp-admin, with a Finish setting up checklist at 2 of 3 done and a Needs review card listing two drafts an agent wrote, each with Preview, Open in editor, Publish and Discard buttons
The plugin dashboard in wp-admin: a setup checklist, then the drafts your agents wrote, waiting for you to preview, publish or discard. Counts, your endpoint and the versions sit further down. A real screenshot of the plugin on a demo site, taken on version 1.9.0.

How it compares with other WordPress MCP plugins.

Several WordPress MCP plugins now start switched off, check permissions on each call and keep a log. What still sets this one apart is the combination, not any single part: all of this plugin’s tools off by default, a dedicated low-privilege agent user the Connection tab can create, no service of ours on the path, and no paid tier. Of the plugins we read on 6 November 2026, none of the others describes all four. Our roundup orders the field by active installs, which puts us last, and each head-to-head says where the other plugin wins.

Common questions about AI agents in WordPress.

The ones people ask first. The full FAQ lives in the docs.

Is it safe to connect an AI agent to my WordPress site?

Yes, as long as the connection is scoped, and scoping is what Agent Abilities for MCP is built around. The agent connects as a real WordPress user and takes that account’s capabilities, so approve the connection while logged in as the account you want the agent to be. Point it at a least-privilege user and it can do no more than that user can do. Every ability stays off until you enable it. Every call to the tools this plugin serves is written to the audit log, denials included. Tools that another plugin publishes straight to the default server of the MCP Adapter are outside these controls, and a Settings switch controls that server.

What happens if the agent tries to delete or break something?

The agent can only do what the WordPress user it connected as can do. Agent Abilities for MCP re-checks that capability before every call. Trash post and Trash page move content to the Trash, where it can be restored, but most Delete tools remove it permanently, so they ship off. The last administrator can never be removed, no matter what the agent asks. Anything the agent is not allowed to do is refused. The refusal is written to the audit log too.

Can I undo what an AI agent changed?

Yes, for edits made with 25 of this plugin’s tools: posts, pages and custom post types, creates, trashes, block edits, terms, custom fields (post and term meta), featured images and SEO fields. The Changes tab keeps up to 30 days and reverts a change if the item hasn’t been edited since, and it refuses a revert that would overwrite a newer edit. It lists field names, never contents, though the earlier values are stored in a table in your site’s database, and when the history is full the oldest values go first. No other tool of this plugin’s can be undone, including permanent deletes, orders, refunds, user accounts, file uploads, WooCommerce products, ACF field values, menus, comments and events, and each of them says so in its response. That includes the WPBakery and Avada tools, but a page edit keeps a WordPress revision when revisions are on, and most of the builder tools return it for restore-revision; their settings tools return the previous values instead, and an imported Avada Studio item is removed by deleting it in Avada’s Library. Tools from other plugins aren’t recorded either. Notifications, webhooks and anything already read are not recalled. Some plugins, GeoDirectory among them, turn revisions off for the whole site; a WP_POST_REVISIONS line in wp-config.php turns them back on. Two optional tools, list-changes and undo-change, let an agent undo its own changes from the last 24 hours, and both start off. History records only while the plugin is active, can be switched off in Settings and is removed on uninstall. WordPress’s export and erase personal data tools cover it.

Can I pause an AI agent without disconnecting it?

Yes. On the Connection tab each approved OAuth connection has Pause and Resume next to Revoke. A paused connection keeps its sign-in and can still list tools, but every tool call returns a message that the connection is paused until you resume it, so nobody has to approve it again. You can also give a connection an end date or a lower rate limit of its own. To stop everything at once, Pause all access refuses every MCP request to the site, for OAuth apps and Application Passwords alike.

Can the agent only save drafts so I approve before anything goes live?

Yes. Turn on Force draft on create in Settings. What the agent creates or asks to publish as a post, page, custom content item, reusable block or WooCommerce product is saved as a draft instead, and a product variation is saved as disabled (private). Edits to a post that is already published still go live, so if you want every change reviewed, connect the agent as a Contributor, who cannot edit published posts at all. The Needs review card on the Dashboard lists those drafts so you can preview, publish or discard them.

Does the agent get admin access?

Only if you connect it as an administrator. The agent takes the capabilities of the account it connects as. Over OAuth that is whichever account approves the connection in the browser, so log in as the dedicated low-privilege user the plugin creates before you approve, and the agent can only do what that user can do. An Application Password works the same way: generate it on that user rather than on your own admin account. Each ability re-checks the capability before it runs. Who may approve an app at all is set under Roles that can connect on the Settings tab, and a new install allows Administrators only.

Can an AI agent issue a refund or change a payment gateway?

Only if you deliberately allow it twice. Ten WooCommerce abilities in Agent Abilities for MCP move money or grant authority. They cover refunds, creating and changing orders, customer updates, payment gateway settings, coupon creation and updates, and tax rate creation and updates. Turning one of those on is not enough by itself. It stays locked until you also switch on a separate high-risk control in Settings. That switch is written to the audit log like any other change. Everything else in the catalog needs only its own toggle.

What happens when another plugin adds new AI tools?

They stay off. Since 1.11.0, when a plugin update or a newly activated plugin adds tools, a “New tools since your last review” card on the Plugins’ own tools tab lists them, and Settings can send a notification. Nothing new reaches an agent through this plugin until you switch it on, and marking a tool as reviewed does not switch it on. The check runs when an administrator loads wp-admin, so a WP-CLI update or an auto-update shows up at the next admin visit. On the first admin load after upgrading, everything already listed counts as reviewed.

Does it send my content to OpenAI, Anthropic, or Google?

No. The plugin connects to no AI provider and sends no telemetry. Your own AI client connects in and calls the abilities you enabled. If you turn on alert emails, they go through WordPress’s own mail to the address you choose and never include call content, a user name or an IP address.

Does it work with ChatGPT?

Yes. Agent Abilities for MCP works with ChatGPT as a custom MCP server. In ChatGPT on the web, open Plugins, add your site’s endpoint as a custom MCP server, and approve it once over OAuth. ChatGPT then calls the abilities you enabled as the scoped user you pick. OpenAI offers full MCP, write tools included, on Business and Enterprise/Edu plans, and Pro can connect with read and fetch tools only. Claude.ai, Claude Desktop, and Manus connect the same way, through a custom connector, and Claude Desktop shares the one Claude.ai uses rather than running a bridge on your machine. Claude Code, Cursor, VS Code, Windsurf, and Gemini CLI connect as well.

How do I connect an AI agent to my WordPress site?

Install Agent Abilities for MCP, then open the first-run Quick Connect wizard on the plugin page. The wizard puts the whole first connection on one screen. For a hosted client like ChatGPT or Claude, turn on OAuth and copy the site endpoint. For a client that runs on your own machine, create a dedicated agent user and generate an application password. Either way, finish by switching on content reads. Nothing this plugin serves is exposed until you enable it. The wizard never turns on a delete.

Can I manage multiple WordPress sites from ChatGPT or Claude?

Yes. Install the plugin on each site and add each site’s endpoint to your AI client as its own connector. The agent can then work across your sites from one place, and each site keeps its own user, its own allowed tools and its own audit log.

New on the blog and in the Prompt Library.

From the blog

From the Prompt Library

All postsAll recipes

Install it, keep it all off, and switch on one ability at a time.

Nothing runs until you choose it, and every call to its tools shows up in your Activity Log.

Free on WordPress.org, with no paid tier waiting behind it. See how undo works.