37 bridged tools from eight plugins worked when called but never appeared in the tool list. 1.7.8 lists them, names the source namespace in bridged errors, and fixes three result bugs.
Read moreThe Abilities tab had a main Save button and four more on its cards, and pressing only the main one lost edits to exposed meta keys and content types. 1.7.7 has one button.
Read moreBefore 1.7.6, an OAuth token for the MCP endpoint also signed its holder in on other parts of a WordPress site. What it reached, what changed, and what else shipped.
Read moreWordPress hands back the same value for "there is nothing here" and "the question could not be asked". Nine review rounds on version 1.7.5 kept finding the same bug wearing a different costume, including one where a failed consent lookup read as consent granted.
Read moreThe per-connection allowlist arrived in 1.7.4 as a textarea you typed ability slugs into. In 1.7.5 it is a searchable checkbox picker with real role and client selects, so narrowing what one connection can reach no longer depends on spelling.
Read moreSend an agent to edit an Elementor page and the write used to report success while the visible page stayed exactly as it was. In 1.7.4 it refuses instead, and says which builder owns the page.
Read moreThe first release since launch that changed the ability catalog. Five new integrations, two new core abilities, a write guard for page builder pages, and an allowlist that only ever narrows.
Read moreOn 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 moreThe sign-in screen now identifies a connector by its verified redirect host rather than the name it sends, and the Activity Log records every MCP request that reaches WordPress.
Read moreA default of ours stopped ChatGPT connecting. Also fixed: a mangled redirect address, a session ID handed back for a session that was never written, and a tool list that could exhaust memory.
Read morePoint 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 moreMost 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 moreA fix-only release: WPML sweeps now read every configured language, tool discovery matches permission, and a broken discovery route resolves again.
Read moreVersion 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 moreVersion 1.7.0 closes an OAuth revocation race, stops a payment gateway credential leak, and fixes four ACF writes that were destroying content.
Read moreIn 1.7.0 an agent asked to set an image's alt text hit a bare permission denied and had no way to know a different tool would have worked. Here is why, and what changed.
Read moreFive 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 moreA 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 moreAn 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 moreThe 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 moreThree 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 moreManaging 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 moreIn 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 moreVersion 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 moreTwo 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 moreA 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 moreIn 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 moreIn 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 moreVersion 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 moreCopy-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 moreLet 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 moreLet an AI agent write your titles, meta descriptions, noindex flags, and social images across Yoast, Rank Math, and AIOSEO, under full governance.
Read moreHow contract tests against real vendor code and write verification keep an AI agent's WordPress tools honest and trustworthy.
Read moreIn 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 moreA 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