Authoring MCP Server

Connect AI tools to Documentation.AI for read/write documentation workflows, authoring, review, and deployment tasks.

Overview

Use the Authoring MCP Server to let an MCP-compatible AI client work directly in your documentation project. It provides read and write access for editing pages, updating documentation.json, managing branches, reviewing changes, and deploying through your normal documentation workflow.

This is the server your documentation team connects to. It is Documentation.AI's endpoint, and it uses your workspace permissions. The server your readers connect to in order to search your published docs is the Reader MCP Server.

The Authoring MCP Server and the Documentation.AI REST API cover largely the same documentation-management tasks, but they expose them through different interfaces. Choose the Authoring MCP Server when you want an MCP-compatible AI client or agent to do the work for you, and choose the REST API when you want direct HTTP access from your own scripts, services, or CI/CD. For endpoint details, see the API reference overview.

Use this server for authoring and documentation operations through an MCP client. If you need direct REST API access instead, use the Documentation.AI REST API. If your AI tool only needs read-only access to published content, use the Reader MCP Server.

This server uses your Documentation.AI account permissions and can modify live documentation. Connect it only to AI tools and workflows you trust, and use branches when you want review before deployment.

Setup and authentication

Connect your MCP client to the server URL, then sign in with OAuth 2.1 through your Documentation.AI account.

https://api.documentationai.app/mcp

Start the connection from your MCP client

Open your MCP-compatible client and add the Authoring MCP Server using the server URL above. For per-client configuration, see Configure your MCP client.

Your client discovers the OAuth endpoints and opens a browser window for authentication.

Approve access

After you approve the consent screen, the client receives tokens and completes the connection.

A successful connection shows the server as available in your MCP client.

No API keys or manual token management are required.

Configure your MCP client

{
  "mcpServers": {
    "documentationai": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.documentationai.app/mcp",
        "--timeout",
        "120000"
      ]
    }
  }
}

{
  "mcpServers": {
    "documentationai": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.documentationai.app/mcp",
        "--timeout",
        "120000"
      ]
    }
  }
}

For Claude Desktop, add the configuration to claude_desktop_config.json. For Cursor, add it to your mcp.json. For Claude Code, run the command from your terminal, or add the same JSON to a .mcp.json file in your project root.

The npx mcp-remote configuration bridges clients that cannot add a remote MCP server directly. If your client supports remote or HTTP servers natively, you can point it at the server URL and let it handle the OAuth flow itself.

Select a project

After authentication, select the documentation project you want to work on before using any documentation tools.

List available projects

Call list_projects to see every organization and documentation project your account can access.

Set the active project

Call select_project with the organization and documentation IDs returned by list_projects.

Run read or write operations

Once a project is selected, use the tools below to inspect files, update content, manage branches, and deploy changes.

A successful selection makes the current project available for subsequent tool calls in the same session.

You can switch projects at any time by calling select_project again.

Tools and capabilities

The server exposes project selection, read, and write tools for repository workflows.

Project selection

These tools are available immediately after authentication.

Read tools

Available to all roles, including viewer, editor, and admin. Use these when your workflow needs repository context, source files, or project configuration.

Write tools

Require editor or admin role. These tools make the Authoring MCP Server the right choice for authoring, automation, and deployment workflows.

Example workflow

This is the branch-and-review pattern the tools are designed around. Ask your MCP client to carry out the sequence, and it calls the tools in order.

Select the project

Call list_projects, then select_project with the organization and documentation IDs you want to work in.

Create a working branch

Call create_branch with a name such as update-api-guides. Working on a branch keeps the deployment branch untouched while changes are in progress.

Make your edits

Call update_documentation with your file operations, a commit message, and branch set to your working branch. Each call commits up to 20 operations at once, so group related edits together.

Review the changes

Use get_page and list_pages against the same branch to confirm the result, and list_pull_requests to check open reviews.

Merge and deploy

Call merge_branches with base set to your deployment branch and head set to your working branch. Merging into the deployment branch triggers a live deployment.

To edit without deploying, stop after the review step and leave the branch open. Nothing reaches your live documentation until a merge into the deployment branch happens.

Resources

The MCP server also exposes reference resources for automation and authoring workflows:

Resource	URI	Description
Components Reference	`docs://components`	All available MDX components with props and usage examples
Config Schema	`docs://config-schema`	Complete `documentation.json` schema reference
Project Config	`docs://project/config`	The current project's `documentation.json`

Role-based access

Your access level is determined by your role in the selected organization.

Role	Permissions
Viewer	Read tools only: `get_site_config`, `list_pages`, `get_page`, `search_documentation`
Editor	Read and write tools, including `update_documentation`, branch operations, merges, and pull requests
Admin	All tools

If your team wants an AI client to read documentation without any ability to change it, use the Reader MCP Server instead.

Troubleshooting

Reader MCP Server

MCP Servers overview

Was this page helpful?