Zed Agent MCP web search / 2026
How to Add Web Search to Zed with MCP in 2026
Zed’s Agent Panel supports remote MCP tools alongside its code-aware features. Webstractor is useful when a development question needs sources beyond the repository: current documentation, a standards page, or a public issue. The integration is a single URL in `context_servers`, with tool approval remaining under Zed’s agent permissions.
Quick answer
Add a remote context server in Zed’s AI settings.
Open Settings → AI → MCP Servers, choose Add Remote Server, and use the Webstractor endpoint. A green status dot confirms the connection. Mention webstractor in your first Agent Panel request to test search_web.
{
"context_servers": {
"webstractor": {
"url": "https://webstractor.com/mcp"
}
}
}Before you begin
What you need
- A current Zed release
- An AI provider configured in Zed
- Access to Agent settings
Step-by-step setup
Connect Webstractor to Zed
Add a remote MCP server
Open Settings → AI → MCP Servers, click Add Server, and choose Add Remote Server. Enter webstractor as the name and the hosted endpoint as the URL. No Authorization header is required for public access.
Review the generated settings entry
Zed writes remote MCP servers under `context_servers`. You can inspect or edit it with the `zed: open settings file` action.
{
"context_servers": {
"webstractor": {
"url": "https://webstractor.com/mcp"
}
}
}Wait for green, then test
The dot beside webstractor should turn green and its tooltip should say Server is active. Open the Agent Panel and make the desired tool and output explicit.
# Paste into the Zed Agent Panel
Use the webstractor search_web tool to find the current Zed MCP documentation. Return 5 results with URLs, then explain which source is official.Agent behavior
Make source use auditable
Zed notes that tool calling varies by model. Mentioning the server name helps the agent select its MCP tools. For stricter behavior, create an Agent profile that enables Webstractor tools and disables conflicting built-in discovery tools.
Keep the returned links in the conversation. When a result looks promising, ask the agent to extract that URL before making a code recommendation.
- Name webstractor in the first prompt
- Require URLs in research output
- Use a dedicated profile for repeatable tool selection
Troubleshooting
Use the status dot and tool permissions
A non-green indicator exposes a tooltip with the connection state. Recheck the URL and remove accidental headers. If the server is active but calls do not happen, inspect `agent.tool_permissions.default`; `deny` blocks MCP tools.
In Zed v0.224.0 and later, the default can be confirm, allow, or deny. Keep confirm while validating the integration so each search is visible before it runs.
Available data
What you can extract
- Ordered public source links and snippets
- Site, locale, country, and PDF filters
- Markdown or schema-v1 JSON result feeds
- A maximum result count chosen per prompt
AI workflows
Where normalized data helps
- Research an API while editing its integration
- Find official docs without leaving the Agent Panel
- Locate standards or public specifications
- Gather source links for a technical decision record
Boundaries
Public data only
- The model can choose not to call a tool unless the prompt is explicit.
- Search results should be followed by extraction when full evidence is needed.
- Only public web information is in scope.
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
Does Zed connect to remote MCP servers directly?
Yes. Add Remote Server accepts a URL and writes it under context_servers.
Why does Zed ask me to approve the search?
The default agent tool permission is confirm. You can change it globally or configure a granular `mcp:webstractor:search_web` rule.
Why is the server active but unused?
Tool choice depends on the model. Mention the webstractor server and search_web tool explicitly or create a focused Agent profile.
Ready to try it?