Write docs with Claude Code

AI coding agents like Claude Code can draft entire documentation pages — but only well if they know Doxbrix's real blocks and your conventions. This guide sets up the skill (workflow instructions) and the MCP server (live block catalog + validation) so an agent authors docs that use real components and pass validation on the first try.

Who this is for

Developers using Claude Code, Cursor, or another MCP-capable AI agent who want to generate and maintain Doxbrix documentation from their editor.

Prerequisites

• A Doxbrix docs-as-code project.
• The dxb CLI installed, with a token.
• Claude Code (or another MCP client).

How it works

Without these, an agent guesses at component names and invents tags that fail lint. With them, it reads the real catalog and validates before writing.

Step 1 — Add the Doxbrix skill

dxb init scaffolds the authoring skill into your project by default:

.claude/
└─ skills/
   └─ doxbrix-authoring/
      ├─ SKILL.md              # the authoring workflow
      └─ blocks-cheatsheet.md  # fallback block reference

Commit it so every contributor's agent follows the same conventions. The skill instructs the agent to: confirm it's a Doxbrix project, learn the blocks, plan structure, write .mdx, validate, and hand off — without ever publishing on its own.

Step 2 — Connect the MCP server

Add the Doxbrix MCP to your agent's configuration:

{
  "mcpServers": {
    "doxbrix": {
      "command": "npx",
      "args": ["-y", "@doxbrix/mcp"],
      "env": {
        "DOXBRIX_TOKEN": "<your-token>",
        "DOXBRIX_PROJECT": "my-docs"
      }
    }
  }
}

Restart the agent. It should now list Doxbrix tools (list_components, validate_mdx, create_page, …). See Connect the MCP.

Step 3 — Ask the agent to write

Write a new guide at guides/integrations/webhooks explaining how to configure
webhooks. Use Steps for the setup, a CodeGroup for the payload examples, and a
Warning about retry behavior. Match the voice of our existing guides.

The agent will: