n8n MCP URL extraction / 2026
How to Fetch Web Pages in n8n with Webstractor MCP in 2026
An n8n AI Agent can read a URL as part of a workflow without hand-building extraction requests. The MCP Client Tool exposes a documented, typed tool to the model, while Webstractor returns clean page content. The workflow still needs guardrails: validate the incoming URL, tell the agent when to use the reader, and preserve the source in whatever system receives the result.
Quick answer
Let the AI Agent read one public URL through its Tool port.
Connect MCP Client Tool to an n8n AI Agent, use HTTP Streamable with the Webstractor endpoint, and expose only extract_public_url. Map a validated public URL into the prompt, request Markdown, and store the returned source URL with downstream output.
MCP Client Tool
Server Transport: HTTP Streamable
Endpoint: https://webstractor.com/mcp
Authentication: None
Tools to Include: Selected → extract_public_urlBefore you begin
What you need
- Current n8n with MCP Client Tool
- AI Agent and Chat Model nodes
- A trigger that supplies one public URL
Step-by-step setup
Connect Webstractor to n8n
Prepare and validate the URL input
Use a Form, Webhook, or upstream node to provide one absolute public HTTP or HTTPS URL. Validate its shape before it reaches the Agent; do not concatenate credentials or private data into query strings.
Configure the MCP reader
Connect MCP Client Tool to the AI Agent’s Tool port. Select HTTP Streamable, use no authentication, and include only extract_public_url. This prevents the same agent from choosing search when the workflow already has a URL.
serverTransport: HTTP Streamable
endpoint: https://webstractor.com/mcp
authentication: None
toolsToInclude: Selected
selectedTools:
- extract_public_urlTell the agent how to handle the page
Reference the incoming URL in the prompt, choose Markdown for reading, and ask for the source URL in structured downstream output. Test with a stable public page before using variable production input.
Use webstractor extract_public_url to read this public URL as Markdown:
{{ $json.url }}
Focus on: {{ $json.focus || 'the main content' }}
Return a concise summary and the canonical source URL.Production design
Keep extraction and business actions on separate branches
Let the MCP reader return evidence before nodes send email, update a CRM, or publish content. A review or validation node between those stages prevents a malformed or irrelevant page from triggering an irreversible action.
If a workflow needs stable routing fields, instruct the tool to return JSON. If a model only needs to read and summarize, Markdown avoids an unnecessary parsing layer.
- Fetch one page
- Validate the result and source
- Then perform downstream actions
Troubleshooting
Inspect node wiring, transport, and item behavior
The MCP Client Tool must connect through the AI Agent’s Tool port, not the main execution line. Choose HTTP Streamable for the /mcp endpoint and Authentication: None.
n8n sub-node expressions resolve against the first input item. If a batch contains multiple URLs, split or loop items deliberately instead of assuming the sub-node will receive a different URL for each item.
Available data
What you can extract
- Markdown from a supplied public URL
- Focused content for a short topic
- Schema-v1 JSON for typed branching
- Canonical source metadata
AI workflows
Where normalized data helps
- Summarize a URL submitted through a form
- Read a documentation link from a ticket
- Prepare source-linked content briefs
- Normalize public product or article pages before routing
Boundaries
Public data only
- The MCP tool reads remote public URLs, not n8n binary uploads.
- Validate caller-supplied fields before placing them in an agent prompt.
- Private pages, crawling, and authenticated sessions are outside this workflow.
webstractor.com does not bypass CAPTCHAs, login walls, paywalls, access controls, or regional restrictions. Review the source’s terms and applicable law before collecting or reusing data.
Common questions
n8n and Webstractor FAQ
Can MCP Client Tool fetch a local PDF file from n8n?
No. extract_public_url reads public URLs. PDF binary upload is available through the Webstractor API, not this MCP URL tool.
Should the n8n agent receive all tools?
No. A known-URL workflow should expose extract_public_url alone unless another explicit capability is required.
How do I process multiple URLs?
Loop or split items explicitly. n8n sub-node expressions resolve against the first input item, so do not assume automatic per-item substitution.
Ready to try it?