API Reference (MCP Tool)

The CRAWL4AI_MCP server exposes a single, powerful tool to your AI assistant: crawl.

Here is the complete list of parameters that your AI can use when calling this tool.

Core Parameters

Parameter	Type	Default	Description
`url`	`string`	(Required)	The fully qualified URL to crawl.
`max_depth`	`integer`	`2`	Maximum crawling depth. Defines how many links deep the crawler should go.
`include_external`	`boolean`	`false`	Whether to include external links in the crawl.
`return_content`	`boolean`	`true`	If true, returns the extracted content directly in the MCP response text.

Parameter	Type	Default	Description
`magic`	`boolean`	`false`	Enable magic mode to bypass anti-bots (e.g., Cloudflare) and simulate a real browser.
`wait_for_selector`	`string`	`null`	CSS selector to wait for before extracting content. Crucial for SPA and React/Vue apps.
`delay_before_return_html`	`number`	`null`	Delay in seconds to wait before extracting HTML (useful for heavy JS pages with slow animations).

Parameter	Type	Default	Description
`css_selector`	`string`	`null`	Specific CSS selector to extract only targeted elements from the page (e.g., `main .article`).
`js_code`	`string`	`null`	Custom JavaScript code to execute on the page before extraction. (Requires server env var `CRAWL4AI_MCP_ALLOW_JS_ENV=true`)

Parameter	Type	Default	Description
`session_id`	`string`	`null`	Persistent session identifier to keep cookies and browser state across multiple requests (great for authenticated sites).

Parameter	Type	Default	Description
`verbose`	`boolean`	`true`	Enable verbose logging output on the server side.
`output_file`	`string`	`null`	Custom path to the output Markdown file. If not provided, it generates a unique name in the `crawl_results` folder.

When Claude or Cline uses this tool, it sends a JSON payload similar to this:

{
  "url": "https://example.com/dashboard",
  "magic": true,
  "wait_for_selector": ".dashboard-loaded",
  "session_id": "auth-session-xyz123"
}