LM Studio MCP URL extraction guide
How to Fetch and Read Web Pages in LM Studio with MCP
Pasting a URL into a local-model chat does not guarantee the model can read that page. A remote extraction tool makes the retrieval step explicit and returns the page as compact model-ready context. In LM Studio, the same MCP configuration works across the local models you test.
Quick answer
Give your local model a focused public-page reader.
Install the Webstractor remote server through LM Studio’s mcp.json, then ask the model to call extract_public_url with an exact HTTPS URL. Request Markdown for summarization or JSON when you need stable fields such as title, source, author, date, and semantic type.
{
"mcpServers": {
"webstractor": {
"url": "https://webstractor.com/mcp"
}
}
}Before you begin
What you need
- LM Studio 0.3.17 or newer
- A tool-calling model loaded in chat
- An ordinary public HTTPS URL to test
Step-by-step setup
Connect Webstractor to LM Studio
Choose a stable public test page
Start with a page that does not require login or interaction. A documentation or Wikipedia article is easier to validate than a personalized application screen.
Edit mcp.json from the Program tab
Open Program → Install → Edit mcp.json. If you already use other integrations, add Webstractor inside the existing mcpServers object.
Register the remote endpoint
Save the following configuration. It connects directly over HTTP and does not launch a local MCP process.
{
"mcpServers": {
"webstractor": {
"url": "https://webstractor.com/mcp"
}
}
}Verify extract_public_url is available
Inspect the integration’s tool list in LM Studio. A successful connection exposes extract_public_url and the related discovery tools.
Test the page-reading path
Prompt: “Call extract_public_url for https://en.wikipedia.org/wiki/Model_Context_Protocol in Markdown. Summarize the definition and history, and cite only the returned source.” Check that LM Studio shows the tool invocation before the answer.
Output choice
Use Markdown for reading and JSON for automation
Markdown preserves headings and readable hierarchy while removing interface noise, which makes it a strong default for a summarization prompt. JSON is more useful when a script consumes the result or when you need an explicit semantic entity type.
Do not ask for both formats just to compare them. Identical extraction inputs share the underlying cached result, but one representation is normally enough for a single task.
Focused reading
Narrow a long page before it reaches the model
When the question concerns one part of a large page, tell the model to pass a short focus such as “pricing” or “authentication.” This reduces irrelevant page text and leaves more of LM Studio’s context window for analysis.
- Name the exact public URL
- State the topic to focus on
- Ask for Markdown
- Require a source-linked answer
Troubleshooting
Separate connection failures from page failures
If no tools appear, validate mcp.json and reload the integration. If the tool appears but one URL fails, test a simple public article next; the original page may require authentication, block access, or expose no readable text.
A model that invents page content without a visible tool call is a model-routing issue. Tell it to call extract_public_url before answering.
Available data
What you can extract
- Readable Markdown from an ordinary public URL
- Normalized JSON for supported products, posts, profiles, feeds, and documents
- Canonical source URL and public metadata
- Focused extraction for a requested page topic
AI workflows
Where normalized data helps
- Summarize release notes with a link back to the source
- Compare pricing pages without pasting HTML
- Turn an article into clean RAG input
- Read supported public product or social pages
Boundaries
Public data only
- The tool reads one public URL; it does not crawl a whole site.
- Login walls, paywalls, CAPTCHAs, private networks, and private pages remain unavailable.
- Web content can contain misleading instructions, so treat retrieved text as untrusted data.
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
LM Studio and Webstractor FAQ
Can LM Studio read any URL after this setup?
It can request supported public HTTP or HTTPS pages. Private, authenticated, paywalled, or otherwise inaccessible content is not fetched.
Which format is best for a local model?
Markdown is usually best for direct reading and summarization. Choose JSON when a program needs typed fields.
Can it extract an entire website?
No. extract_public_url handles one URL per tool call; it is intentionally not a crawler.
Ready to try it?