Plasmic MCP. Manage projects, pages & components via AI.

You're gonna run your whole headless content flow straight from your chat window, no browser context switching required. This server gives your AI agent direct access to Plasmic’s backend functions, letting you manage entire site drafts and deploy changes right from your code editor or terminal. You get full control over the project lifecycle using simple commands.

Project Configuration and Global Control

You can change overall project settings or metadata for the whole site with update_project. This tool lets you trigger global updates that affect how Plasmic renders content across multiple pages. For listing available work, your agent pulls all defined projects when you reference the capability to list them.

Data Retrieval and Structure Inspection

Need to know what data's sitting in a CMS collection? You use query_cms_items to fetch filtered lists of records from any designated Plasmic database table based on criteria you specify. If you just need a rough count, count_cms_items tells you the total number of entries in a table. To understand how a component is built, your agent calls get_model, which returns the full JSON structure (SDUI) detailing all internal components and their properties for any project element.

Content Management: Drafting and Modification

When you're working on drafts, you manage content records using standard CRUD operations. To start fresh, your agent adds a brand-new item or draft record to the CMS via create_cms_item. If existing content needs tweaking—say, changing some body copy or swapping out an image in a draft—you call update_cms_item to modify specific details of an existing entry.

You can permanently remove records using delete_cms_item, which takes an item from any specified CMS table.

Publishing and Rendering Output

Drafting is only half the battle; you gotta get it live. To move a draft content item into a visible state, your agent executes publish_cms_item. This function makes sure that piece of content shows up on the actual site. For final output, if you need to test exactly what a component looks like before deployment, render_html generates and returns the clean HTML code for any specific component instance.

The entire process operates by having your AI client treat these tools like native functions; it calls them when you ask it to perform an action—like 'Give me the latest copy for the pricing page' or 'Check if the contact form is ready.' When the agent executes a tool call, it returns the raw data or confirmation back to you in plain language.

You’re essentially running core CMS operations without leaving your IDE. This gives your AI client power over Plasmic's backend functions, letting you manage content and project states entirely through natural conversation.