Review what your AI agent wrote before it goes live
Make an AI agent save drafts for you to approve. Turn on Force draft on create, use the Needs review card, and see which edits to live posts it does not hold.
Read moreWhat 1.7.8 changes: bridged tools now show up in the tool list
37 bridged tools from eight plugins worked when called but never appeared in the tool list. 1.7.8 lists them, names the source in errors, and fixes three bugs.
Read moreWhat 1.7.7 changes: one Save changes button on the Abilities tab
The Abilities tab had a main Save button plus four on its cards, and pressing only the main one lost edits to meta keys and content types. 1.7.7 has one button.
Read moreWhat 1.7.6 changes: an OAuth token now works on the MCP route only
Before 1.7.6, an OAuth token for the MCP endpoint also signed its holder in elsewhere on a WordPress site. What it reached, what changed, and what else shipped.
Read moreAbsent and unknown are not the same answer
WordPress returns one value for both 'nothing here' and 'could not ask'. Nine review rounds on 1.7.5 kept finding that bug, once in a consent check.
Read moreWhat 1.7.5 adds: picking abilities instead of typing them
The per-connection allowlist was a typed textarea in 1.7.4. In 1.7.5 it is a searchable checkbox picker with role and client selects, so no more spelling slugs.
Read moreA write that changed nothing now refuses by name
Ask an agent to edit an Elementor page and the write used to report success while the page stayed the same. In 1.7.4 it refuses and names the builder.
Read moreWhat 1.7.4 adds: the catalog moves from 153 to 179
The first release that changed the ability catalog: five new integrations, two core abilities, a page builder write guard, and an allowlist that only narrows.
Read moreWhat we fixed: a switch that would not turn off
On sites with a persistent object cache, read-only mode and the high-risk lock could refuse to change, and the activity log recorded the change as done anyway.
Read moreWhat 1.7.2 adds: a consent screen that shows who is asking
The sign-in screen now identifies a connector by its verified redirect host, and the Activity Log records requests that reach WordPress and fail.
Read moreWhat we fixed in the 1.7.2 connection release
A default of ours stopped ChatGPT connecting. Also fixed: a mangled redirect address, a session ID for a session never written, and a tool list that ate memory.
Read moreA read-only look at your site and its plugins
Point an agent at a WordPress site for a plain inventory of the basics and the installed plugins, without handing it any power to change a thing.
Read moreThe abilities we deliberately left out
Most of the pitch is about what you can switch on. This post is about the floor under all of it: what an agent cannot do here, no matter what you enable.
Read moreWhat we fixed in the 1.7.1 correctness release
A fix-only release: WPML sweeps now read every configured language, tool discovery matches permission, and a broken discovery route resolves again.
Read moreWhat 1.7.0 adds: Enable all writes, and a fuller audit log
Version 1.7.0 adds a one-click way to enable a section's ordinary writes, deliberately leaving deletes and high-risk abilities locked, plus a fuller audit log.
Read moreWhat we fixed in the 1.7.0 correctness release
Version 1.7.0 closes an OAuth revocation race, stops a payment gateway credential leak, and fixes four ACF writes that were destroying content.
Read moreWhat we fixed: setting alt text said permission denied, for everyone
In 1.7.0 an agent setting an image's alt text hit a bare permission denied, with no hint that another tool would work. Why that happened, and what changed.
Read moreAI agent WordPress security risks
Five real risks of an AI agent on WordPress, deletion, data leakage, injected content, and credential exposure, mapped to the control that limits each one.
Read moreWordPress AI prompt injection, contained
A hidden instruction in a comment can fool an AI agent on WordPress. What indirect prompt injection is, and how governance contains it when detection can't.
Read moreLimit an AI agent to Editor, not Admin
An AI agent takes on whatever WordPress user connects it. Here is why Editor beats Administrator as that account, and how to scope the connection on purpose.
Read moreWordPress MCP Adapter vs a governed layer
The WordPress MCP Adapter is the official building block for AI agents. Agent Abilities for MCP adds a governed catalog, admin UI, and audit log on top of it.
Read moreA WordPress AI audit log you can trust
Three 1.6.2 fixes make the plugin's audit log worth trusting: it records the true cause of a block, covers every path that changes state, and stays readable.
Read moreGuard WooCommerce customer data from AI agents
Managing a store and editing a user are different powers. Here is why customer data belongs behind the capability that actually governs editing user accounts.
Read moreWe said the audit log stored nothing. That was not quite true.
In 1.6.2 we corrected a claim about the audit log. It keeps identifier-only values by design, never free-text content, and here is why that line matters.
Read moreWhat we fixed in the 1.6.2 hardening release
Version 1.6.2 corrects a claim about our own audit log, tightens who can edit customer data, and closes gaps in what the plugin records and sanitizes.
Read moreHigh-risk lock and read-only mode, explained
Two controls decide what an AI agent is even offered on your WordPress site, before any per-call permission check runs: the high-risk lock and read-only mode.
Read moreWhat we fixed in the 1.4.1 standards release
A user-reported OAuth bug and six related fixes bring error responses, discovery, and tool status codes in line with what MCP and OAuth clients expect.
Read moreWhat we fixed: media reads reached past the uploader
In 1.4.3 an author-level user could read the entire WordPress media library, not just their own uploads. Here is the bug, the fix, and what else shipped.
Read moreWhat we fixed when a draft post went live
In 1.4.2 we fixed a bug where asking for a draft post published it live, two related permission gaps, and two WooCommerce order and product update bugs.
Read moreQuick Connect: one-screen setup for a WordPress MCP server
Version 1.4.0 adds a one-screen Quick Connect wizard that gets an agent connected fast, without loosening the off by default governance underneath.
Read moreFirst prompts to try on your WordPress MCP server
Copy-paste prompts for a freshly connected WordPress MCP server: safe reads, a careful draft write, and what a refusal looks like when an agent overreaches.
Read moreAI agents for membership, subscription, and LMS stores
Let an AI agent manage customers and orders on a membership, subscription, or LMS WooCommerce store, safely and audited, now that custom roles are visible.
Read moreHand your SEO metadata to an AI agent, safely
Let an AI agent write your titles, meta descriptions, noindex flags, and social images across Yoast, Rank Math, and AIOSEO, under full governance.
Read moreReliable agent tools: contract tests and verified writes
How contract tests against real vendor code and write verification keep an AI agent's WordPress tools honest and trustworthy.
Read moreWhat we fixed in the 1.3.0 security release
In 1.3.0 we turned OAuth off by default and fixed real security gaps in our own plugin, patched in the open. Here is what changed.
Read moreOff by default: how the governance model works
A walk through the governance model behind Agent Abilities for MCP, covering least privilege, off by default, capability gating, and an honest audit log.
Read more