Authoring MCP Server

Overview

The Authoring MCP Server lets an MCP-compatible AI client work in your documentation project. It can read pages, write and reorganize content, add images and files, update documentation.json, manage branches, and publish changes.

This is the server your documentation team connects to, and it uses your role in your Documentation.AI organization. To give readers AI access to your published docs, use the Reader MCP Server. To work over HTTP from your own scripts or CI/CD, use the REST API.

How edits work

• Changes the agent makes to pages appear in the web editor as unpublished changes, the same as a teammate's edits.
• Nothing goes live until the changes are published.
• Publishing a branch includes every unpublished change on it, including your teammates' edits.
• Media is the exception. Your media library is shared by every branch, so importing, replacing, or deleting a file takes effect right away. Replacing or deleting a file changes every page that uses it, including your live site.

Setup and authentication

Add the server URL to your MCP client, then sign in with your Documentation.AI account. No API keys are needed.

https://api.documentation.ai/mcp

Configure your MCP client

claude mcp add --transport http documentation-ai https://api.documentation.ai/mcp

{
  "mcpServers": {
    "documentation-ai": {
      "url": "https://api.documentation.ai/mcp"
    }
  }
}

{
  "servers": {
    "documentation-ai": {
      "type": "http",
      "url": "https://api.documentation.ai/mcp"
    }
  }
}

{
  "mcpServers": {
    "documentation-ai": {
      "command": "npx",
      "args": ["mcp-remote", "https://api.documentation.ai/mcp"]
    }
  }
}

Choose a project and branch

Tell the agent which documentation project and branch to work on. It finds and selects the project with list_projects and select_project, and works on your live branch unless you name a different one.

Example workflow

Describe the task to your AI client in plain language, and it calls the tools for you.

Add images and files

Tell the agent which image or file you want and where it goes. The agent adds the file to your media library, then inserts it into the page with the same snippet the web editor writes.

• If the file is at a public URL, give the agent the URL and the place on the page. For example: Import https://example.com/setup-diagram.png and add it under the Install heading in guides/setup.mdx
• If the file is on your computer, ask the agent for an upload link. Open the link, sign in if asked, and drop your files. Then tell the agent you're done, and it finds the files in your media library. The link works for 15 minutes.
• If the file may already be there, ask the agent to find it by name or alt text first, and reuse it instead of importing a second copy.
• If you want a newer version on every page, ask the agent to replace the file from a URL. The URL stays the same, so pages update without edits.

For step-by-step instructions, limits, and supported file types, see Media library.

Tools reference

The server provides 32 tools. Read-only tools never change your documentation. Destructive tools can overwrite or discard content, including your teammates' unpublished edits, so most clients ask you to confirm them first. Only tools that add something new, such as creating a page or importing a file, go through without a prompt.

Most tools also accept organizationId and documentationId to work on a specific project, and branch to work on a specific branch. Tools that return lists accept limit and cursor to page through results.

Projects

Authoring reference

Read content

Edit content

Review and publish

Branches

Media

Files belong to your documentation project, not to a branch, so these tools do not take branch. Each file gets its own result, so one failed file does not stop the others.

Resources

The server also provides these reference resources:

Role-based access

Your role in your Documentation.AI organization decides which tools you can use.

If you want an AI client to read your published documentation without being able to change it, use the Reader MCP Server instead.

Troubleshooting

Related resources