Prepr MCP server

From this guide, you’ll learn how to set up the Prepr MCP (Model Context Protocol) to help editors manage content from an AI agent or built-in AI tools, like Notion AI.

Introduction

Connect AI clients to your Prepr environment with a Model Context Protocol  server at https://mcp.prepr.io/mcp. The Prepr MCP server helps editors and teams find content, create and update localized items, review publication state, update workflow, assign work, patch supported fields, and publish or unpublish items directly from AI tools.

The Prepr MCP server (Beta release)

The Prepr MCP server gives AI clients a safe, structured way to work with content in your Prepr environment.

With the MCP server, an AI client can trigger the following actions:

• Search for content items across your environment
• Search for assets such as images, videos, audio, and documents
• Upload an image or document asset from a public URL or file path
• Fetch an item with a matching Content item ID
• Create a new localized content item
• Update an existing localized content item
• Publish items now or schedule publishing for later
• Unpublish items
• Change workflow stages
• Assign or unassign items
• Add comments to content items
• Patch supported root fields
• Delete an item or a specific locale with explicit confirmation

Connecting to the Prepr MCP server

To connect the Prepr MCP server, follow the steps below.

1. Request a Prepr MCP token with the permissions you want the AI client to have.

2. Add a remote MCP server to your client.

3. For example, in Notion, click the + New agent menu item and enter a description like Creates and updates Prepr CMS content items from Notion pages, sets the requested stage and assigns them to a user. Once created, go to the Settings, add a connection and choose to add a custom MCP.

   

4. Set the server URL to https://mcp.prepr.io/mcp.

5. Pass your token in the Authorization header.

Safety and limitations

The MCP is designed to help with editorial work, but write actions still need the right context.

Important behavior:

• Write actions depend on your Prepr MCP token permissions
• If you’re connecting to a multi-locale environment, the AI agent should ask for clarification when it’s not clear which locale to create or update a content item
• For a single-locale environment, include the locale in the AI agent’s core instructions.
• If multiple candidate items match the request, the AI agent should ask for clarification before mutating content
• Deleting content always requires explicit confirmation

Current limitations:

• The create_item request creates one locale per call and supports model-level root fields of type Text, Boolean, Integer, Float, Enum, Color, Tags, ContentReference, Asset, and ElementBox
• If you only have a public image or document URL, the model can first use upload_asset and then use the returned asset ID in an Asset field or dynamic content block
• The update_item request uses the same simplified input shape as create_item, but sends a full PUT update for one locale after first merging your changes into the current item
• The patch_items request supports model-level root fields of type Text, Boolean, Integer, Float, Enum, Color, Tags, and ContentReference
• The bulk mutation calls are processed in chunks of up to 25 content item IDs per call

Currently known bugs and PR’s

• Resource needs to be added with payload examples for the dynamic content editor
• Stack field is not yet supported when creating or updating content items.