WordPress.com today expanded its Model Context Protocol integration with write capabilities, allowing AI agents such as Claude, ChatGPT, and Cursor to create, edit, and delete content directly on WordPress.com-hosted sites. The announcement, made on March 20, 2026, by developer advocate Jonathan Bossenger, marks a significant shift for the platform - moving AI agents from passive observers of site data to active participants in full site management.

The update arrives roughly five months after WordPress.com first introduced read-only MCP support in October 2025, which allowed AI tools to query site content, analytics, and settings. Thousands of users connected AI clients to their sites under that initial rollout. According to the March 20 announcement, users consistently pushed for agents that could act on that data, not simply read it. The write capabilities are the direct response.

What the write capabilities actually do

The new functionality adds 19 distinct operations across six content types: posts, pages, comments, categories, tags, and media. Before this update, an AI agent connected to WordPress.com could tell a user what was published on their site - traffic figures, recent drafts, pending comments. Now, the same agent can draft a blog post, categorize it, write a meta description under 160 characters, approve pending comments, build a new taxonomy structure, and fix missing image alt text - all through natural language instructions, without the user opening a dashboard.

According to the support documentation last reviewed on March 16, 2026, the full list of operations available through the MCP tools includes viewing site settings and configuration, checking site statistics and traffic data, searching and retrieving posts, reviewing comments, checking plugin status (for administrators), viewing site users (for administrators), creating and editing posts and pages, managing comments, and organizing categories and tags. The 19 new write tools extend the already-available read tools, so the integration covers both directions of site management within a single connected session.

The write capabilities work through the existing MCP dashboard. According to the announcement, no new software is required on the user's side. The only required steps are enabling MCP at wordpress.com/me/mcp, toggling on the specific write capabilities desired, and connecting an MCP-compatible AI client.

How Claude connects: the official partnership

WordPress.com is listed as an official partner in the Claude Connectors Directory. According to the support guide last reviewed on March 18, 2026, connecting Claude Desktop to WordPress.com requires no manual configuration. The process involves opening Claude Desktop, navigating to Settings, clicking the Connectors tab, browsing connectors, searching for "WordPress.com," selecting the official connector, clicking Connect, and granting Claude access to the account through a standard OAuth authorization screen. After those steps, Claude can access WordPress.com sites through whichever MCP tools the user has enabled.

Anthropic introduced its Integrations feature in May 2025, which opened Claude to remote MCP servers across both web and desktop applications. Before that, MCP support for Claude was limited to local servers accessed only through Claude Desktop. The WordPress.com connector sits within that broader Integrations infrastructure - meaning it is reachable from Claude's web interface as well, not only the desktop application. The Claude support guide specifies that if the connection drops, the troubleshooting step is to click the tools icon in the Claude Desktop message bar and select "Retry Connection" next to the WordPress.com connector, then verify that MCP access is still enabled in account settings.

How ChatGPT connects: developer mode required

The ChatGPT connection path is more involved. According to the support guide last reviewed on March 16, 2026, connecting ChatGPT to WordPress.com requires a paid ChatGPT subscription - specifically Plus, Pro, Team, or Enterprise. The Free and Go plans do not support this feature. Team and Enterprise plan users additionally need a workspace admin to enable "Allow members to develop and sideload custom apps" in the Admin Console before the connection can be established.

The technical setup involves enabling Developer Mode in ChatGPT settings - accessed through the profile icon, Settings, Apps tab, and then Advanced settings - and then creating a custom app. The custom app configuration requires entering the MCP server URL: https://public-api.wordpress.com/wpcom/v2/mcp/v1. After the app is created, a browser window opens to authorize the connection through OAuth 2.1. Each time a new chat session starts in ChatGPT, the user must explicitly select the WordPress.com connector from the app list above the message bar. Without that selection step, ChatGPT does not have access to the site tools in that conversation. If the connection stops responding, the reconnection path is through Settings, the Apps tab, finding the WordPress.com app, and clicking Reconnect.

How other MCP clients connect

The MCP endpoint https://public-api.wordpress.com/wpcom/v2/mcp/v1 is documented for use with any compatible client. According to the MCP support guide, compatible clients include Cursor, VS Code, Gemini, and Perplexity, in addition to Claude and ChatGPT. The OAuth 2.1 authentication flow runs automatically on first connection, guiding users through browser-based authorization without requiring manual credential configuration. This standardized authentication approach reflects the MCP protocol's design: a single integration method that works consistently regardless of which AI client is being used.

The underlying MCP standard - originally developed by Anthropic and later donated to the Linux Foundation - defines how applications provide context to large language models. Anthropic launched the protocol in November 2024, initially with support limited to Claude Desktop through local servers. The protocol's adoption has since extended across dozens of platforms and AI clients. According to the MCP support documentation, the WordPress.com MCP server never shares any data with the AI model unless the user explicitly sends it as part of a request. The server also does not use data from MCP tool interactions to train AI models - data is used only within the scope of the original request.

Design-awareness as a technical distinction

One capability that sets this implementation apart is theme integration. Before generating content, according to the announcement, an AI agent can search the active theme's design data - understanding its colors, fonts, spacing, and block patterns. Content created by the agent then inherits those design properties. When a user changes themes, the output adapts automatically.

This is practically significant for content and marketing teams. A common failure mode of AI-generated web content is visual disconnection from the surrounding site design - formatting that the theme does not support, or ignoring block patterns the designer already built. By scanning theme data before producing output, the agent's drafts are more likely to match what the site actually requires, without needing manual reformatting after the fact. The announcement describes this as one of the most powerful aspects of the write capability rollout.

The full list of operations and what each covers

The support documentation provides a comprehensive view of the kinds of tasks now available through natural language. On the read side, an agent can identify which site gets the most traffic, show recent posts, suggest new blog post topics, or summarize recent comments. On the write side, the documented capabilities break down as follows.

For post and page management: an agent can write and save a draft post in the user's style, create a new page with specified sections, update the title of the most recent post, add an excerpt to a draft, and change the status of a draft from unpublished to live. For comment management: an agent can approve pending comments, reply to specific comments, mark comments from particular authors as spam, and delete comments. For taxonomy management: an agent can create new categories, add subcategories, create tags, and apply tags to specific posts. For media management: an agent can update alt text on recently uploaded images and update captions on named media files. For content cleanup: an agent can move published posts back to draft, or trash all posts that have been in draft status for more than a specified period.

These are not hypothetical capabilities - each maps to a documented prompt the platform explicitly supports as of the March 16, 2026 support guide revision.

Safety architecture: approval, drafts, trash, and permissions

WordPress.com built several constraints into the write system, each addressing a specific category of risk that emerges when autonomous software can modify a live website.

Every change requires explicit user confirmation before it executes. The agent describes what it intends to do and waits for approval. Nothing changes without that sign-off, according to both the announcement and the support documentation. The language in the MCP guide is specific: "Write operations (including delete) will request confirmation before executing the command."

New posts and pages default to draft status. When an agent creates content, it does not go live immediately. Users review the draft before publishing. If an agent updates a post that is already published, it warns the user that the change will be immediately visible to site visitors.

Deletions follow a conservative path. When an agent removes a post, page, comment, or media file, the item moves to the WordPress trash - recoverable for 30 days. Categories and tags are a technical exception: WordPress does not support trashing taxonomy items, so the agent explicitly warns users that those deletions are permanent and requests additional confirmation before proceeding. Both the announcement and the support documentation repeat this distinction, indicating it is a deliberate communication point for users unfamiliar with how WordPress handles taxonomy deletion differently from content deletion.

All agent activity is logged in the site's Activity Log, which users can consult independently or query through the agent itself. According to the announcement, users can ask the agent directly for a list of changes it has made - closing the audit loop within the same conversational interface used to initiate the changes.

WordPress permission roles remain in force

The write capabilities do not override WordPress's built-in user role system. An Editor account connected to an AI agent can create and edit posts but cannot change site settings. A Contributor-level connection can draft posts but cannot publish. According to both the announcement and the MCP support guide, existing access controls carry over automatically - the agent operates within whichever role it is assigned, with no escalation of privileges.

The MCP support guide specifies this explicitly: "MCP respects the user role permissions on your site. For example, an Editor can create and edit posts, but can't change site settings. A Contributor can draft posts but can't publish." This means that organizations assigning limited WordPress roles to team members retain those same limitations when those accounts are used to connect AI clients. The agent cannot do more than the human account behind it is authorized to do.

Users also retain granular control at the individual operation level. Every capability - creating posts, updating media, managing comments - has its own toggle in the MCP settings. Operators can enable only what they need on the specific sites they need it, and disable everything else. The MCP guide also supports disabling all MCP access for individual sites under the "Site-specific MCP settings" section, allowing operators running multiple sites to apply different permission configurations to each.

Availability and supported AI clients

Write capabilities are available on all WordPress.com paid plans as of the March 20 announcement date. The supported AI clients include Claude from Anthropic, Cursor, ChatGPT from OpenAI (Plus, Pro, Team, and Enterprise tiers), and any other application that supports the MCP standard. Gemini, Perplexity, and VS Code are also mentioned in the support documentation as compatible clients.

Context: how this fits the broader MCP landscape

The MCP standard has seen rapid adoption across technology platforms since its launch in November 2024. Google explored MCP integration with its Ads API in July 2025, then released an open-source version in October 2025. Google Analytics launched its own MCP server on July 22, 2025, enabling conversational queries against analytics properties. AppsFlyer launched an MCP tool in July 2025, targeting mobile marketing teams. Amazon Ads opened its MCP Server to open beta on February 2, 2026, connecting advertisers' AI agents to campaign management and reporting APIs through natural language. FreeWheel launched its own MCP server in March 2026, extending the standard into the premium video ad-serving layer.

WordPress.org itself announced four AI building blocks in July 2025, including a PHP AI Client SDK, an Abilities API, an MCP Adapter, and an AI Experiments Plugin - establishing foundational infrastructure for AI features across the open-source WordPress ecosystem. That initiative, led by Automattic's James LePage as Engineering Director of AI, positioned WordPress sites as both MCP servers and clients: capable of exposing capabilities to external AI tools while also connecting to other MCP-compatible systems. The WordPress.com write capabilities announced today represent a consumer-facing product built partly on that foundation and delivered to hosted plan subscribers without requiring any developer configuration.

The WordPress.com integration is notable for the Claude side of the equation as well. Anthropic launched its Integrations feature in May 2025, extending MCP connectivity from local servers to remote servers accessible across Claude's web and desktop applications. HubSpot launched its first CRM connector for Claude in July 2025, establishing a pattern of third-party platforms building official Claude connectors. WordPress.com is now listed in the Claude Connectors Directory as an official partner - a status that simplifies the connection process compared to other MCP clients and reflects a formal partnership between the two companies.

WordPress.com's relationship with AI dates back further still. In July 2024, the platform partnered with Perplexity AI, integrating its publisher network into the Perplexity content discovery engine, with revenue sharing for publishers whose content appeared in AI-generated answers. That deal was primarily about content distribution. The MCP write capabilities announced today represent an entirely different type of AI integration - one where the AI operates as an active agent within the site rather than indexing the site's output for external consumption.

Why this matters for content and marketing teams

For digital marketers, content strategists, and site operators, the practical implications extend beyond individual task automation. The write capabilities change the relationship between an AI assistant and a CMS from a drafting aid to an operational layer.

The alt text audit capability illustrates the scale of what becomes possible. Alt text is both an accessibility requirement under standards such as WCAG and an SEO signal. It is frequently incomplete across sites with large media libraries, because updating it has historically required navigating each media item individually in the WordPress dashboard. An agent that can identify all images missing alt text across a library and suggest descriptions based on filename and attachment context automates a task that would otherwise require hours of manual work across potentially hundreds of files.

The taxonomy management tools address a similar operational bottleneck. Creating a new category structure, adding subcategories, and tagging a defined set of existing posts are tasks that site editors regularly defer, because they require several layers of dashboard navigation. Through natural language, those same operations become a single instruction - and the agent confirms each step before executing.

The comment management capabilities matter for sites with active communities or high comment volumes. Moderating comments, replying on behalf of the site, and removing spam are recurring tasks. An agent with access to these operations can process a backlog through a single session, provided the user confirms each action. The approval requirement, built into the system at the protocol level, means the agent cannot act autonomously - but it does reduce the number of clicks and page loads required to process each item.

The content cleanup functions address the problem of stale content. Drafts that have sat unpublished for more than a year represent a common form of CMS clutter. The ability to instruct an agent to trash all such posts - with confirmation before each deletion - enables site auditing at a scale that is impractical when performed manually.

The design-aware content generation addresses a longer-standing tension in AI content tools: outputs that are textually acceptable but structurally misaligned with the site's layout. Theme-aware generation reduces the gap between what the agent produces and what the site actually requires. The draft default means users still review before anything goes live, preserving editorial control.

The IAB Tech Lab's Agent Registry, which reached 10 entries in March 2026, reflects how quickly MCP-based infrastructure has grown across advertising and media technology. The Trade Desk is testing Claude for campaign creation in a closed beta, placing it alongside Amazon, Google, FreeWheel, and now WordPress.com as platforms building production-grade MCP integrations. For marketing teams running content operations on WordPress.com alongside advertising campaigns on those platforms, a consistent AI agent layer - accessible through the same clients and the same conversational interface - reduces the operational fragmentation that has historically characterized multi-platform digital marketing.

Timeline

Summary

Who: WordPress.com, operated by Automattic, announced the write capabilities on March 20, 2026. The announcement was made by developer advocate Jonathan Bossenger. The update affects all WordPress.com paid plan subscribers who use AI clients compatible with the Model Context Protocol, including Claude (via the official Claude Connectors Directory partnership), ChatGPT (Plus, Pro, Team, and Enterprise plans with Developer Mode enabled), Cursor, VS Code, Gemini, Perplexity, and any other MCP-compatible application. The Claude connection guide was last reviewed March 18, 2026; the ChatGPT and MCP access guides were last reviewed March 16, 2026.

What: WordPress.com expanded its existing MCP integration with write capabilities, adding 19 new operations across six content types - posts, pages, comments, categories, tags, and media. AI agents can now create, edit, organize, and delete site content through natural language instructions, subject to user approval at each step. The system includes theme-aware content generation, draft defaults for new content, trash-based deletion with 30-day recovery for posts and media (with a permanent deletion warning for categories and tags), full activity logging, and enforcement of WordPress user role permissions. The MCP server endpoint is https://public-api.wordpress.com/wpcom/v2/mcp/v1 and uses OAuth 2.1 authentication. ChatGPT connections require Developer Mode and a paid ChatGPT subscription. Claude Desktop connections use the official WordPress.com connector in the Claude Connectors Directory.

When: The write capabilities were announced on March 20, 2026, and were made immediately available on all WordPress.com paid plans. The preceding read-only MCP support had launched in October 2025.

Where: The functionality is accessible through the MCP settings dashboard at wordpress.com/me/mcp and applies across all WordPress.com paid hosting plans. It operates with any MCP-compatible AI client and uses a standardized OAuth 2.1 authentication flow for all non-Claude clients.

Why: According to the announcement, users who connected AI clients under the original read-only MCP integration consistently requested the ability to take action, not just query data. The write capabilities respond to that demand, expanding the agent's role from informational assistant to active site collaborator while maintaining approval gates, draft defaults, trash-based reversibility, and existing WordPress permission controls.

Share this article
The link has been copied!