LobeChat Desktop URL extraction / 2026
How to Fetch Web Pages in LobeChat with Webstractor MCP in 2026
Pasting a link into chat does not guarantee the model reads its current content. Webstractor provides an explicit read-only extraction tool, while LobeChat Desktop provides the assistant and model interface. The result is best for known public pages: documentation, articles, product pages, social posts, and other supported URLs.
Quick answer
Give a LobeChat assistant a public-page reader.
Import the Webstractor bridge as a custom MCP plugin in LobeChat Desktop, enable it for the active assistant, and pass an exact public URL to extract_public_url. Ask for Markdown and a source-linked summary.
{
"mcpServers": {
"webstractor": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
}
}
}Before you begin
What you need
- LobeChat Desktop
- Node.js and npx
- A public URL and a tool-capable model
Step-by-step setup
Connect Webstractor to LobeChat
Open LobeChat’s JSON importer
Navigate to Settings → Default Assistant → Plugin Settings → Custom Plugins → Quick Import JSON Configuration.
Install the Webstractor bridge
Import this entry and enable it for the default assistant. The `-y` option lets npx install the adapter without a confirmation prompt.
{
"mcpServers": {
"webstractor": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
}
}
}Test a known public page
State that the assistant must use the extraction tool. Ask it to preserve the page URL and distinguish source facts from its summary.
# Paste into LobeChat
Use webstractor extract_public_url to read https://lobehub.com/docs as Markdown with focus "self-hosting". Summarize the relevant options and include the source URL.Prompt design
Tell the assistant what to read and why
Include the exact URL, preferred output, and a concise focus topic. “Read this page” is workable, but “extract the installation section as Markdown and make a checklist” produces a result that is easier to verify.
For comparisons, perform one extraction per URL and ask LobeChat to keep separate source headings. This avoids mixing facts from pages that may have different publication dates.
- Pass the canonical page URL
- Use Markdown for a model-readable result
- Request attribution in the final answer
Troubleshooting
Diagnose the adapter before the page
If no Webstractor tools appear, reload the plugin configuration and confirm that Node.js can run npx. If the tools appear but one URL fails, try a stable public page to determine whether the source itself is restricted.
Image-only PDFs do not contain extractable text without OCR. For normal PDFs, use a public PDF URL; uploaded local files belong to the API upload workflow rather than this URL-only MCP tool.
Available data
What you can extract
- Readable Markdown without surrounding page controls
- Optional focused extraction
- Normalized JSON entities for structured tasks
- Source-linked metadata for attribution
AI workflows
Where normalized data helps
- Summarize a linked report
- Read current documentation before answering
- Compare two public product or article pages
- Turn a long page into a focused brief
Boundaries
Public data only
- Only the supplied public URL is read.
- The plugin does not access private sessions or browser cookies.
- This custom-plugin process applies to LobeChat Desktop.
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
LobeChat and Webstractor FAQ
Will LobeChat automatically open every pasted URL?
Not reliably. Ask it explicitly to use webstractor extract_public_url for the link.
Can the tool read pages behind a login?
No. It does not receive browser sessions, cookies, or credentials.
Can it extract only one topic?
Yes. Pass a short focus such as pricing, installation, or limitations.
Ready to try it?