Aller au contenu principal

Author MCP

Pro

The author MCP is a private, token-authenticated server that gives any MCP-capable AI agent the same powers as the in-editor AI agent. Generate a token, drop it into your AI tool's config, and you can manage your whole docs site in natural language.

What you can do

"Update the quickstart with the new 30-second flow." "Scrape acme.com/changelog and add this week's release as a Changelog page." "Find every mention of 'Webhook' and link it to the new Webhooks page." "What's changed in this site's docs since last week? Show me a summary."

Your AI agent reads the current docs, drafts the changes, shows you the diff before anything goes live, and publishes when you say.

Generating a token

Project settings → MCP → Author MCP → Generate token.

  • Give the token a name (e.g. "Claude Code on my laptop")
  • Click Create
  • Copy the token (shown once — won't be visible again)

The token format is docsio_auth_<random>. Store it like an API key.

Where to paste it

Each MCP tool has a config file. Setup snippets for the most common tools are in Setup snippets. The pattern:

{
  "mcpServers": {
    "<your-project-name>-author": {
      "type": "http",
      "url": "https://docsio.co/api/mcp/author/<your-token>"
    }
  }
}

Restart your AI tool. It picks up the new MCP server automatically.

What the agent sees

  • All your pages (read access)
  • Project metadata (name, description, custom domain, page count)
  • Recent unpublished changes
  • A 16-tool surface for editing and publishing

The 16 tools

Read your docs:

  • get_project_info — name, description, live URL, page count.
  • list_pages — every page with title, slug, category, word count.
  • list_project_structure — full file tree.
  • read_page — full markdown of a page.
  • search — full-text search.

Edit your docs:

  • write_page — create or update a page.
  • create_page — explicit new page (manages frontmatter).
  • delete_page — remove a page.
  • rename_page — change a page's slug, fix internal links.
  • upload_asset — upload an image to your static/img/ folder.
  • update_project_info — change name or description.

Bring in outside content:

  • scrape_url — read a single URL as clean markdown.
  • map_site — list every URL on a site (useful for planning).
  • batch_scrape — many URLs at once.

Review and publish:

  • get_unpublished_changes — diff between your draft and the last published version.
  • publish — build and deploy.
  • get_publish_status — live status of an in-flight publish.

Safety

  • Writes always stay in draft until you say publish. The agent can write 50 files; nothing goes live until publish is called explicitly.
  • Tokens are scoped to one project. A leaked token only affects that project, not your account.
  • Tokens are revocable. Revoke any token instantly from settings; it stops working within seconds.
  • Rate limits. Publish is rate-limited to 5 per hour per project to prevent runaway loops.

Multiple tokens

You can have multiple active tokens per project at once — for example:

  • One on your laptop in Claude Code
  • One on your home machine in Cursor
  • One in Codex on a CI runner

Up to 10 active tokens per project. Manage and revoke individually in settings.

When to use Author MCP vs the editor

  • Editor: visual, has live preview, designed for review-as-you-go.
  • Author MCP: text-driven, faster for batch operations, lets you stay in your IDE.

Most users do both. "Let Claude Code rewrite the API reference" (MCP) → "open in editor to spot-check" (visual).

What's logged

Every tool call is logged to a per-project audit log: timestamp, tool, outcome, file changed. Visible in project settings → MCP → Author MCP → Recent activity. Helpful when you want to know what your agent did while you weren't watching.

Free vs Pro

Author MCP is Pro-only.

  • Setup snippets — exact config for Claude Code, Cursor, VS Code, Codex, Antigravity.
  • Reader MCP — the public version of MCP, for your end-users.