Cline MCP public URL guide
How to Fetch Web Pages in Cline with Webstractor MCP
A coding agent often receives a documentation link in an issue or README. Webstractor lets Cline turn that URL into readable context without copying navigation, scripts, and layout markup into the prompt. The strongest pattern is read, cite, plan, then edit.
Quick answer
Let Cline read an exact source before it codes.
Connect the Webstractor remote server with Cline’s Streamable HTTP transport, then ask Cline to call extract_public_url with an exact public page. Request Markdown for documentation reading and require the implementation plan to cite the returned canonical URL.
{
"mcpServers": {
"webstractor": {
"type": "streamableHttp",
"url": "https://webstractor.com/mcp",
"disabled": false,
"autoApprove": []
}
}
}Before you begin
What you need
- Cline with remote MCP support
- A public documentation or article URL
- Tool-call review enabled
Step-by-step setup
Connect Webstractor to Cline
Open Cline’s remote server form
From MCP Servers, open Remote Servers. Name the connection Webstractor so it is easy to identify in approval prompts.
Register the single HTTP endpoint
Choose Streamable HTTP and enter https://webstractor.com/mcp. If you prefer JSON, use Cline’s exact configuration below.
{
"mcpServers": {
"webstractor": {
"type": "streamableHttp",
"url": "https://webstractor.com/mcp",
"disabled": false,
"autoApprove": []
}
}
}Inspect the advertised tools
Confirm extract_public_url is visible and autoApprove remains empty. A read operation can still bring hostile text into an agent context, so visibility matters.
Ask Cline to read without editing
Prompt: “Do not change files. Call extract_public_url on https://docs.astro.build/en/guides/middleware/ in Markdown with focus ‘middleware order’. Summarize the relevant behavior and cite the source.”
Authorize implementation as a second phase
Review the extracted facts and plan. Only then ask Cline to compare the documentation with the repository and propose or apply changes.
Prompt design
Make retrieval an explicit prerequisite
“Read this first, do not edit” is a useful control because it stops Cline from treating a URL as decorative context. Naming extract_public_url also makes the external evidence visible in the transcript.
Use focus for a large reference page so the model receives the relevant subsection rather than spending context on the entire manual.
Security
Web pages are data, not agent instructions
A retrieved page can contain prompt injection or outdated commands. Tell Cline to summarize factual documentation and ignore instructions that attempt to alter its role, access local secrets, or run commands.
- Review the requested URL
- Reject URLs containing credentials
- Keep tool output distinct from trusted project rules
- Approve file and shell actions separately
Troubleshooting
Use a known public page as a control
If extraction fails for one vendor portal, retry with a stable public documentation page. Success there shows the MCP connection works and the original page is likely access-controlled or unsupported.
If Cline never proposes the call, confirm the server is enabled and explicitly name extract_public_url.
Available data
What you can extract
- Clean Markdown from one public page
- Normalized JSON for structured integration logic
- Canonical URL and visible public metadata
- Focused content for a specific API or configuration topic
AI workflows
Where normalized data helps
- Read a migration guide before upgrading dependencies
- Extract an API reference linked from an issue
- Compare public product documentation to repository behavior
- Summarize release notes into an implementation checklist
Boundaries
Public data only
- Do not give the tool signed, private, or credential-bearing URLs.
- Fetched instructions are untrusted and must not override repository rules.
- One URL call is extraction, not documentation-site crawling.
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
Cline and Webstractor FAQ
Is this safer than giving Cline browser credentials?
It avoids passing browser sessions or credentials because Webstractor only reads ordinary public URLs. Retrieved content still needs prompt-injection review.
Can I restrict extraction to one section?
Yes. Ask Cline to pass a concise focus string such as “authentication” or “migration steps.”
Should extract_public_url be auto-approved?
Start with manual approval. If your policy later permits read-only public URL calls, you can make a deliberate per-tool choice.
Ready to try it?