WordPress.com has brought about a significant change in how websites are managed by allowing the Claude tool with MCP support They can not only read, but also write and modify content directly on sites hosted on their platform. From now on, tools like Claude, ChatGPT or Cursor They can literally become editorial assistants capable of taking care of a good part of the day-to-day work, although always with the human being in charge.
With this move, Automattic, the company behind WordPress.com, transforms its MCP server on a write-protected gateway to websites, and not just in a window for consulting data. For those who manage blogs, digital media, or corporate projects in Spain and the rest of Europe, this innovation represents another step towards the automation of editorial tasks, but it also opens a debate about the impact that the increase in AI generated content in the quality and trust of the users.
What can AI agents do now within WordPress.com
The new version of the WordPress.com MCP server incorporates writing and active management skills on the site, moving from read-only access to full interaction with the content. These agents are no longer limited to listing entries or displaying statistics: they are capable of create, edit and organize virtually all the key elements of a website.
According to Automattic, the update adds 19 additional operations distributed across six large blocksThis includes posts, pages, comments, categories, tags, and a media library. In practice, this means that a user can ask their agent, with a natural language command, to write an article, generate a landing page, or reorganize the content structure of their site.
Among the most notable actions, AI agents can write and publish posts, build pages using the active theme's block patterns, modify existing text, approve or reject comments, clean spam, group posts under new categories or tags, and Correct the alternative text for the images to improve accessibility and SEO.
A key aspect is the so-called “design awareness” of the topicBefore creating a new page, the agent analyzes the site's design system—colors, fonts, spacing, and block patterns—so that the generated content visually matches the existing style. This way, the AI-generated pages don't appear as extraneous elements, but rather integrated into the same aesthetic as the rest of the project.
Regarding the publication, New entries are generated as drafts by defaultThis allows the site owner or editorial team to review, adjust, or discard the content before it goes live. If the AI suggests changes to already published content, the system explicitly states that the changes will be visible immediately, so the user is fully aware of the impact.
A model of human supervision and permission control
Beyond the technical aspects, WordPress.com has placed emphasis on the trust design and human oversightInstead of allowing agents to act autonomously, every significant action first goes through an approval filter: the agent explains what it intends to do and waits for explicit confirmation from the user before executing it.
This means that, if AI proposes delete a page, modify key text, or clear commentsThe site owner will see a detailed summary of the transaction and must manually approve it. In a context where failures of automated tools have generated embarrassing headlines, this architecture seeks to reduce risks and maintain editorial control in human hands.
The WordPress.com user roles are strictly adhered to.An Editor can create and modify entries, but not certain global settings; a Contributor can generate drafts, but not publish them. AI agents inherit these limitations, so they cannot bypass existing editorial governance.
Deletion operations are also handled carefully. Posts, pages, comments, and media files are first sent to the trash and can be recovered for 30 days. In the case of categories and tags, which WordPress cannot send to the trash, the system introduces an extra confirmation step, warning that it is a permanent removal.
Furthermore, All actions performed through MCP are recorded in the Activity Log of the site. This allows for auditing what was done, when, and through which agent, something especially relevant for projects with distributed teams, agencies, or communication departments that must account for the changes introduced.
How to activate and configure AI agents on WordPress.com
To take advantage of these new capabilities, WordPress.com customers should go to the dedicated dashboard in wordpress.com/mcpFrom there, the MCP server is activated and the granular settings are defined. operations that are to be allowed on each site: for example, you can allow the creation of drafts but block the deletion of content, or facilitate the management of comments while prohibiting changes to categories and tags.
The connection between WordPress.com and the agents is made through OAuth 2.1The authorization standard that allows you to securely grant and revoke permissions. The WordPress.com MCP server is optional and It is not activated by default, something the company emphasizes as a control element for site administrators.
Once the server is enabled, the owner can connect your preferred AI client provided it is compatible with MCP. Among the most prominent are Claude (Anthropic), ChatGPT (OpenAI), and Cursor, although the MCP ecosystem is open and continues to grow with new tools that can be integrated into this workflow.
From the AI client, the user interacts with their bot just as they would in a regular chat: writes or dictates instructions in natural language And the agent, with access to the site's context (content, structure, analytics, and relevant settings), suggests specific actions and changes to the website.
For those working on multiple projects, this per-site configuration allows adjust the level of AI autonomy Depending on the sensitivity of each website: there may be a testing environment where the agent has more freedom to experiment and, at the same time, a main corporate site where its capabilities are limited to low-criticality tasks.
MCP as infrastructure for the era of AI agents
The WordPress.com movement is part of the evolution of Model Context Protocol as a de facto standard to connect AI agents with real-world applications. MCP, originally powered by Anthropic and then donated to the Linux Foundation, it has expanded to include thousands of servers available, offering a unified way for models to access and act upon external systems.
In this context, WordPress.com has been laying the groundwork for some time. read-only MCP integration It arrived in October 2025, allowing agents to view content, analytics, and configurations without being able to modify them. In early 2026, OAuth 2.1 support was added, and an official connector for Claude was released, also initially limited to read access.
The writing capabilities being launched now are, in a way, the logical step of a phased roadmapFirst, the query is opened, then authentication is secured, and once the security requirements are met, the action on the site is enabled. In parallel, other Automattic products, such as WooCommerce or the Beeper messaging appThey have been incorporating their own MCP implementations.
For the WordPress ecosystem as a whole, the company is working on the WordPress MCP AdapterDesigned to bring similar capabilities to self-hosted installations (those using WordPress.org on their own or third-party servers), this adapter is progressing towards potential integration into the WordPress core, which would extend AI agent logic beyond WordPress.com and its hosting plans.
All of this occurs on a foundation of enormous scope: WordPress powers over 43% of the world's websitesand WordPress.com represents a significant portion of that ecosystem, with figures around 20.000 billion monthly visits and 409 million unique visitorsThe ability of AI agents to write and organize content at this scale has implications that go far beyond a simple productivity improvement.
Impact on content creators, media and SEO in Spain and Europe
For professionals in content marketing, digital media and SEO In Spain and the rest of Europe, the integration of agents via MCP presents both opportunities and risks. On the one hand, small newsrooms and teams can automate repetitive tasks —summaries, rewrites, tags, alt text, standard comment responses— and dedicate more time to strategy, research, and analysis; a useful resource to get started is the Artificial intelligence glossary for creatives.
On the other hand, the fact that such a high percentage of the web can be updated almost on autopilot with AI assistance opens the door to a notable increase in automatically generated contentThis raises reasonable concerns about the saturation of poorly differentiated articles, the loss of editorial voice, and users' ability to distinguish between in-depth texts and purely algorithmic pieces.
In the field of search engine optimization, the possibility of AI Optimize metadata, reorganize category structures, and correct descriptions It can help improve visibility, especially for websites that have been plagued by years of poorly maintained content. However, there is also a risk that poorly supervised strategies can lead to low-quality practices, something that European search engines are increasingly monitoring. For approaches and best practices related to SEO and AI, see a guide on digital marketing and artificial intelligence.
On the regulatory front, the massive influx of automated content coincides with debates in the European Union on the transparency in the use of AI and responsibility for the published information. WordPress.com's design of the system with human approval, activity logging, and voluntary activation is a clear attempt to fit into this regulatory environment, although ultimate responsibility will still rest with site owners.
Ultimately, for agencies, specialized blogs, digital newspapers, and European corporate projects, the question is no longer so much whether to use AI in content management, but how to integrate it without losing editorial controlThe tool is available, but how it is used will make the difference between improving processes or generating more noise.
WordPress.com is thus at the center of a silent but profound transformation: with an MCP server capable of giving write access to AI agents and an ecosystem that covers a large part of the web, content management is entering a phase where collaboration between humans and machines can accelerate daily work, provided that those who manage the sites maintain a clear criterion about what to delegate, what to review and what to leave exclusively in human hands.
