Coauthor

Coauthor is a set of tools, instructions, and skills that let AI agents read and edit the CKEditor 5 documents of your users. Your agent connects to them through an MCP server in CKEditor Cloud Services. Your backend can call the same tools over REST, without an AI model.

Coauthor runs CKEditor 5 on the server, with the editor bundle that you upload to Cloud Services. It checks every edit in this editor before it applies the edit. The agent’s edits reach the document as direct changes or as Track Changes suggestions.

Coauthor runs no AI model of its own. You choose the agent and its model.

To learn what Coauthor is for and when to use it, see Coauthor in the CKEditor 5 documentation. This guide shows how to set up Coauthor and connect your agent to it.

Without Coauthor, your agent gets the HTML through your API, rewrites it with its own tools, and saves it back. With Coauthor, the agent sends each edit to Coauthor, which checks it in your editor and applies it as a separate change. People can review each change.

• Editor compatibility. Coauthor tells the agent which features your editor has, such as charts or footnotes, and gives it a short guide for each one. Coauthor validates each edit using CKEditor 5 and your editor bundle before applying it.
• No lost content. Coauthor rejects edits that would result in lost content, such as a chart, a style, or a comment that your editor would drop. It tells the agent what would be lost, so the agent can correct the edit and try again. An agent that rewrites the HTML does not learn what your editor drops.
• Review by users. The agent’s edits arrive as separate, granular changes, not as a rewritten document. Users can review and accept or reject the agent’s suggested changes.
• Editing at the same time. Users can keep writing while the agent works, and the agent’s edits merge with their changes. Saving a rewritten document could overwrite changes made by other users.
• Limited access. Permissions limit what the agent can do. An agent that saves through your API can save anything that the API allows.

To learn how Coauthor handles an edit, see How Coauthor works.

Coauthor works with any agent framework and any model, including small ones. Your agent needs no library to use it.

• Any MCP client. Your agent connects to one MCP address. Claude Code, Cursor, and agents that you build with the Claude Agent SDK, the OpenAI Agents SDK, the Vercel AI SDK, LangChain, or your own code can all use it.
• Frontier models. Your agent can use the leading models from OpenAI, Anthropic, and Google, such as GPT-6 Sol, Claude Opus 5.5, Claude Sonnet 5.5, and Gemini 4 Argon.
• Small models. We build and test Coauthor on small, low-cost models, such as Claude Haiku 5.5, GPT-6 Luna, and DeepSeek V4.1 Flash.
• Open-weight models. Your agent can use a model that you host yourself, such as gpt-oss-120b, Qwen3.8-27B, Mistral Large 4, or Kimi K3.1 Pro.
• Your infrastructure. The agent may run on a user’s computer or in a cloud.

To connect your agent, see Connect an agent.

CKEditor AI is our agent, built into your editor. Coauthor lets your own agent edit the documents, wherever the agent runs. You can use CKEditor AI and Coauthor with the same documents.

• CKEditor Cloud Services (SaaS): in Early Access. We turn on Coauthor for an environment that you already have, or for a trial environment. Contact us to join.
• On-premises: coming soon.

• Quick start: edit a document in a trial environment, first with Claude Code, then with your own code.
• How Coauthor works: follow one job from the agent’s first call to the change in the document.
• Set up your environment: connect your own agent.
• Security and compliance: how Coauthor handles and protects your document data.