Fresh out of the oven

Prismic MCP can now publish a release, and only when someone asks it to. You say it in words, something like "publish the August promo release," and it goes through only if you already have publishing rights in the repository. The agent can't do anything you couldn't do yourself.

Nothing is scheduled and nothing runs on its own. The agent never decides a release is ready, it still can't delete anything (documents are archived, not deleted), and it still stages its work as drafts. You decide what goes live.

What this unblocks

Until now, the MCP did the whole job and then stopped one step short. It staged everything as drafts, and someone had to open Prismic to push the release live. That's fine for most content work, but it meant every automated job ended in a manual step, and some of the things customers asked us for weren't possible at all. A bulk edit, a localization pass, or a migration can now run start to finish in the tool you're already working in.

Also in the Migration API

You can publish a release from the Migration API too, so it works the same whether you're in an AI tool or in your own code.

We're working on a clearer way to confirm a publish from inside the MCP. More on that soon.

The Prismic MCP docs cover setup, connection, and the full list of tools.

Most content work still happens in two places at once. You draft in ChatGPT or Claude, then copy it into Prismic by hand, one document at a time. Renaming a feature across thirty landing pages, or localizing a campaign across markets, can eat hours of clicking with no record of what changed.

Prismic MCP closes that gap. It's the official MCP server that connects AI tools to your Prismic content, so the work happens inside the agent you already have open. Ask Claude or ChatGPT to make the change, and it builds the result as drafts in a release. Your team reviews and publishes the way it works today.

What you can do

Teams are already using Prismic MCP to:

• bulk-edit content across many pages at once, like renaming a product or swapping a CTA across every feature and landing page
• localize a campaign across markets, instead of copy-pasting between ChatGPT and Prismic
• roll out a messaging change everywhere it appears, like updating a plan name, a price, or a key stat across the whole site
• migrate content into Prismic from another platform, by giving the agent your content and a field mapping

Safe by design

You can give this to your team without worrying about the AI making a mess. Prismic MCP can't delete your content and can't publish on its own. Every change is saved as a draft, nothing is ever deleted (only archived), and a person is always in charge of deciding what goes live. The safety guarantees are built in, not optional settings you have to remember to switch on.

Free on every plan, set up in minutes

Prismic MCP is currently free on every plan, and any Prismic user can turn it on for their repository with an in-app button. No developer, no procurement step, no waiting on your CSM. The docs walk through setup, connection, and the full list of tools. It works with any MCP-compatible client, including Claude, Claude Code, ChatGPT, Codex, and Cursor.

MCP and the Prismic CLI work together

If you write code, the Prismic CLI is the companion piece. The CLI handles project structure, so AI tools can create types, slices, and scaffolding. Prismic MCP handles content, so AI tools can read, edit, and populate content in a repository you've already built. While you're building, that means you can populate content and test the pages you develop straight from your agent, in Claude, Cursor, or ChatGPT, instead of hand-filling the editor. Used together, the CLI and MCP take a project from setup to published content inside the same AI workflow.

Note: The old prismic/prismic-mcp-server for AI-assisted slice coding is deprecated and replaced by the Prismic CLI. If you land on the old GitHub repo, head to the Prismic CLI instead.

AI coding agents are already great at writing code, reading designs, and building components, and Prismic fits naturally into that workflow. The one piece agents couldn't reach on their own was the Prismic-specific work that lived in the UI or in model files, like starting a project, modeling content, or syncing types.

The Prismic CLI gives agents a reliable way to do that work from the terminal, an interface they already know how to use. Ask Claude, Cursor, ChatGPT, or Codex to build with Prismic, and the agent runs official commands to set up the project, model content, and get you to a working local preview. Fewer setup dead ends, less manual clicking, a clearer path from prompt to running project.

What you can do

Ask your agent to:

• initialize a Prismic project in your codebase, with the SDKs and config wired up
• model content and sync types without hand-editing model files
• go from prompt to a working local preview of your first page

How to try it

Point your agent at the docs and tell it what you want. For example:

Set up Prismic in this Next.js project and get me to a working local preview of my first page. Use the Prismic CLI and follow the official docs.

CLI and MCP work together

The CLI and the Prismic MCP cover different jobs. The CLI is the developer surface for project setup, content modeling, sync, and local development. The Prismic MCP is the content surface, where agents read, edit, and manage content. So while you're building, you can use the MCP to populate content and test the pages you develop straight from your agent, instead of hand-filling the editor. Build the project with the CLI, then run content operations with the MCP.

Note: The Prismic CLI replaces the old prismic/prismic-mcp-server for AI-assisted slice coding. If you land on the old GitHub repo, head to the Prismic CLI instead.