Zed Agent URL extraction / 2026
How to Fetch Web Pages in Zed with Webstractor MCP in 2026
A link in a comment or issue does not automatically become useful model context. Zed can call a remote MCP reader that converts the public page into readable content and keeps the source URL attached. This works especially well when you ask the Agent Panel to compare external documentation with the current buffer or repository.
Quick answer
Turn a pasted public link into Agent Panel context.
Configure Webstractor as a remote context server, confirm its green status indicator, and ask Zed to use extract_public_url on the exact link. Markdown suits code reasoning; JSON suits structured transformations.
{
"context_servers": {
"webstractor": { "url": "https://webstractor.com/mcp" }
}
}Before you begin
What you need
- Current Zed with Agent Panel access
- Configured AI provider
- A public page to read
Step-by-step setup
Connect Webstractor to Zed
Connect the remote reader
From Settings → AI → MCP Servers, choose Add Remote Server. Use the name webstractor and the /mcp endpoint. Zed also lets you make the same edit directly in settings.json.
{
"context_servers": {
"webstractor": {
"url": "https://webstractor.com/mcp"
}
}
}Check server and permission state
Wait for the green Server is active indicator. Keep the default confirm permission while testing, or add a granular rule for `mcp:webstractor:extract_public_url` if your workflow needs automatic reads.
Read a source next to the code
Open the relevant project and ask for a focused read followed by a comparison. The URL in the answer is the audit trail.
# Paste into the Zed Agent Panel
Use webstractor extract_public_url to read https://zed.dev/docs/ai/mcp with focus "tool permissions". Compare the documented behavior with this project’s settings and cite the page.Context design
Combine the page with the current buffer deliberately
Ask Zed to distinguish facts read from the URL from conclusions drawn from repository code. This reduces the chance that a source statement and a model inference are presented as the same thing.
For large pages, a short focus topic reduces unrelated content. Remove focus when the task is an overall summary or when a missing subsection may change the conclusion.
- Name the exact page
- Identify the repository files to compare
- Request the source URL in the conclusion
Troubleshooting
A green server does not guarantee every page is readable
The green dot confirms MCP connectivity. A per-page failure can still mean the source is private, unavailable, or contains no readable public text. Test a stable documentation page to isolate the problem.
If Zed chooses another tool, mention `webstractor extract_public_url` explicitly or configure an Agent profile in which the Webstractor reader is the only external fetch tool.
Available data
What you can extract
- Readable Markdown for the Agent Panel
- Optional focus on a named section
- Structured JSON when fields matter
- Canonical URL and public metadata
AI workflows
Where normalized data helps
- Check a linked migration guide against current code
- Read a standards page while implementing a parser
- Summarize a public issue beside the affected file
- Compare two vendors’ public API documentation
Boundaries
Public data only
- The server cannot read workspace files; Zed supplies those separately.
- Public URL extraction does not authenticate to websites.
- One extraction request covers one URL, not a full crawl.
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
Zed and Webstractor FAQ
Can Zed compare fetched content with my repository?
Yes. Zed supplies workspace context and Webstractor supplies the public page; ask the agent to label which conclusions come from each.
Can I auto-approve only URL extraction?
Yes. Zed supports granular MCP permission keys such as `mcp:webstractor:extract_public_url`.
Does Webstractor read localhost documentation?
No. It accepts public HTTP or HTTPS URLs, not private network or local addresses.
Ready to try it?