webstractor

LobeChat Desktop web search / 2026

How to Add Web Search to LobeChat with MCP in 2026

LobeChat’s assistant experience becomes more useful for research when it can return source URLs instead of answering only from model knowledge. Its documented Desktop workflow imports MCP server JSON through Custom Plugins. The configuration below uses a small local bridge to the hosted Webstractor server, so it requires Node.js but no locally hosted search service.

20264 min readFor agent builders

Quick answer

Import Webstractor into LobeChat Desktop as a custom MCP plugin.

LobeChat Desktop documents custom MCP import under the default assistant’s plugin settings. Import the JSON configuration, enable the plugin for the assistant, and ask it to use search_web. The local mcp-remote bridge connects LobeChat’s stdio plugin slot to the hosted endpoint.

MCP configuration
{
  "mcpServers": {
    "webstractor": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
    }
  }
}

Before you begin

What you need

  • LobeChat Desktop
  • Node.js with npx available
  • A tool-capable model selected for the assistant

Step-by-step setup

Connect Webstractor to LobeChat

01

Open custom plugin import

In LobeChat Desktop, go to Settings → Default Assistant → Plugin Settings → Custom Plugins, then choose Quick Import JSON Configuration. This is the installation path documented in the LobeHub MCP catalog.

02

Import the hosted-server bridge

Paste the configuration and install it. `mcp-remote` adapts the hosted HTTP endpoint to the local stdio process expected by this import format.

MCP import JSON
{
  "mcpServers": {
    "webstractor": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
    }
  }
}
03

Enable the plugin and prove search ran

Enable webstractor for the assistant, start a new conversation, and request URLs. A source list makes a tool-backed answer distinguishable from an ungrounded response.

Validation prompt
# Paste into LobeChat
Use webstractor search_web to find the official LobeChat MCP plugin documentation. Return five URLs and identify which one documents installation.

Assistant setup

Keep discovery separate from page reading

Configure the assistant to use search_web when the user has a question but no URL. Once it finds a relevant page, have it call extract_public_url for full content. This two-step policy keeps search snippets from being mistaken for complete evidence.

Require citations in the response and choose a small limit. Broad searches create more context but rarely produce a better chat answer.

  • No URL → search_web
  • Known URL → extract_public_url
  • Need full evidence → extract the chosen result

Troubleshooting

Check Desktop, Node.js, and plugin state

The custom import instructions apply to the Desktop app. If the process fails to start, run `npx -y mcp-remote https://webstractor.com/mcp` in a terminal to confirm Node.js and network access.

If the plugin is installed but unused, enable it for the active assistant and mention the webstractor tool by name in the validation prompt.

01

Available data

What you can extract

  • Ordered public web links and snippets
  • Result limits and optional site restriction
  • Locale-aware discovery
  • PDF-only source searches when requested
02

AI workflows

Where normalized data helps

  • Build a source-linked research assistant
  • Find official product or developer documentation
  • Collect candidate sources before drafting
  • Search a specific public hostname from chat

Boundaries

Public data only

  • This guide targets LobeChat Desktop’s documented custom-plugin flow.
  • The npx bridge runs locally and therefore requires Node.js.
  • Search accesses public sources only and does not reuse browser logins.

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

Why does this setup use npx?

LobeChat Desktop’s documented custom MCP import uses local process configuration. mcp-remote bridges that process to Webstractor’s hosted HTTP endpoint.

Does the bridge host a search server on my Mac?

No. It is a transport adapter; retrieval still happens through the hosted Webstractor MCP service.

Can LobeChat search one domain?

Yes. Ask it to pass that hostname in the search_web site parameter.

Ready to try it?

Turn a public URL into useful context.

Connect LobeChat Desktop Explore Webstractor MCP