Recently, I came across a new Block MCP plugin from GravityKit that was highlighted by The Gutenberg Times. In an article, GravityKit introduced Block MCP, an open-source WordPress MCP server designed to let AI agents edit WordPress content at the block level without breaking Gutenberg block structure.
Why GravityKit Built Block MCP
According to the article, GravityKit developed Block MCP after encountering challenges using AI tools to create and update content on its own website.
Many existing WordPress MCP solutions rely on the standard WordPress REST API, treating a post as a single HTML document. While that approach works for publishing content, GravityKit found that AI-generated edits could result in invalid Gutenberg blocks and broken page layouts when content was reopened in the WordPress editor.
As Casey Burridge explains:
Block MCP is the WordPress MCP server we built to let AI agents edit posts at the block level without breaking the block structure. It’s now open source and available on GitHub.
Instead of editing an entire post as one HTML document, Block MCP exposes content as a structured tree of individual Gutenberg blocks.
What Makes Block MCP Different?
The article describes Block MCP as consisting of two components:
- A WordPress plugin that exposes Gutenberg content as structured blocks.
- An MCP server that enables AI agents to interact with those blocks.
According to GravityKit, this allows AI tools to work with individual content elements rather than rewriting an entire page for every change.
The company says AI agents can:
- Edit, insert, move, duplicate, and delete blocks.
- Perform batch updates across multiple blocks.
- Maintain references through persistent block IDs.
- Roll back changes using WordPress revisions.
- Update Yoast SEO metadata alongside content edits.
The goal is to enable more precise content editing while preserving Gutenberg’s underlying structure.
How It Compares with the WordPress REST API
A major theme of the article is the distinction between block-level editing and traditional REST API workflows.
GravityKit notes that most WordPress automation tools ultimately interact with content as a single HTML field. As a result, even a minor update may require reading, modifying, and rewriting an entire post.
By contrast, Block MCP allows AI agents to work directly with individual blocks. The company argues this makes targeted edits easier and reduces the risk of invalid block markup.
Examples cited in the article include:
- Updating a single heading without rewriting the rest of the page.
- Performing multiple edits within a single revision.
- Searching content at the block level.
- Preserving Gutenberg block markup during edits.
- Preventing deprecated or unsupported blocks from being introduced.
Benchmark Results
GravityKit also shares benchmark testing comparing Block MCP with other WordPress MCP solutions.
According to the company, Block MCP achieved a 100 percent success rate in a simple heading-conversion test conducted across three Claude models, while competing solutions produced lower success rates. Additional tests involving more complex structural edits, such as moving blocks and modifying tables, also reportedly favored Block MCP.
These results come from GravityKit’s own testing and should be viewed in that context, but they provide insight into the approach the company has taken.
An Emerging Pattern in WordPress AI Tooling
One of the more interesting aspects of the article is its focus on architecture rather than features.
GravityKit argues that the key distinction among WordPress MCP solutions is whether they operate on raw post content or on Gutenberg’s block structure. The company positions Block MCP as a response to the growing need for AI systems that can safely edit existing WordPress pages without disrupting layouts, patterns, and block configurations.
As AI-assisted publishing workflows continue to evolve, tools that understand WordPress content at the block level may become increasingly important.
Getting Started
According to GravityKit, Block MCP is open source and available under the MIT license. The project can be installed as a WordPress plugin and connected to MCP-compatible AI clients using standard WordPress Application Passwords for authentication.
Acknowledgement: This summary was generated using AI writing tools as part of my AI Explorations process. For more information, please refer to the original article linked above.
Related Resources
- Block MCP | GitHub
- Introducing Block MCP: the WordPress MCP we built because nothing else worked | Gravity Kit
- Using the REST API | WordPress Developer resources