PlugWiki
Web Search MCP logo

Web Search MCP Server: Free Local Web Search for Local LLMs

CommunityLocal

Gives a local model web search and page reading with no API key, using search engines and a headless browser on your own machine.

This is a community server by mrkrsl, built for local models and tested by its author with LM Studio and LibreChat. It runs on your computer and needs no account or key. It searches Bing, Brave and DuckDuckGo directly and can open the result pages in a headless browser.

You install it from a release download, not from a package manager. You build it once with npm and install the Playwright browsers.

Good for

  • People running local models in LM Studio or LibreChat who want web search without a paid search API
  • Developers who want current web pages read in full and passed to a model
  • Anyone who prefers search requests to go out from their own computer rather than through a key-based service

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install Web Search MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Web Search MCP server to Claude Code. Run this command:
    
    claude mcp add web-search -- node /path/to/web-search-mcp/dist/index.js
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js 18+, npm, Playwright browsers.

Read-only mode: Not needed: the tools only read.

Install first: download the latest release zip from the Releases page, unzip it, then run npm install, npx playwright install and npm run build in that folder. Replace /path/to/web-search-mcp with the real folder; on Windows use double backslashes in JSON. The README documents LM Studio and LibreChat; this config for your app is our adaptation of the same command.

Built from the package data we hold. Not yet tested by us.

Tools

  • Searches the web and reads the top pages in full

    Runs a web search, opens the top results (1 to 10) and extracts the full text of each page. It tries Bing and Brave in a headless browser first and falls back to DuckDuckGo. The server tells the assistant this is the most complete option.

    Try: “Search the web for the latest Playwright release notes and summarise them”

    Network · full-web-search

  • Searches the web and returns only snippets

    Runs the same search but returns just the title, link and snippet for each result without opening the pages. It is quicker and lighter on context.

    Try: “Give me quick search results for "sqlite wal mode"”

    Network · get-web-search-summaries

  • Reads one web page you name

    Opens a single URL and returns its main text with menus and ads removed. You can cap the length with maxContentLength.

    Try: “Read https://example.com/docs/setup and list the steps”

    Network · get-single-web-page-content

Read from the source code

Permissions

low risk · 2 things to watch

Low risk

The tools only read search results and web pages, and change nothing on your machine. Pages can contain instructions aimed at the model, so review what it does after reading them.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • It opens pages the assistant picks

    Search results and URLs are opened in a browser on your machine, and page text goes to the model. A page can contain instructions aimed at the model, so review what it does next. github.com

  • It automates search engine pages

    It drives Bing and Brave with Playwright browsers and falls back to DuckDuckGo over plain requests. Search engines may block or slow this, and the README tells you to try another network if that happens. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyNot needed: none of the 3 tools writes or deletes anything.
Talks toBing, Brave and DuckDuckGo for searches. Any web page it opens from search results or a URL you give it
Sign-in and keysNo credentials are needed.

Questions

Is the Web Search MCP server official?

No. It is a community project by mrkrsl. It is not made by Bing, Brave, DuckDuckGo or LM Studio.

Does it need an API key?

No. It searches the engines directly from your computer with no key or account.

Which apps does it work with?

The author says it was built and tested with LM Studio and LibreChat and has not been tested with other MCP clients. Any app that can start a local node program may work.

How do I install it?

It is not on a package manager. Download the latest release zip, run the install, Playwright browser and build steps in the folder, then point your app at the built file. The install box shows the steps.

Is it still maintained?

The newest release is v0.3.2 from August 2025, and the last commit is from the same month. Check the repository before you rely on it.

Which models work best with it?

The README says recent tool-capable models work best. It lists Qwen3 and Gemma 3 as the best results and warns that older Llama and Deepseek versions can be erratic.

Does it send my searches to a service?

Searches go straight from your computer to Bing, Brave or DuckDuckGo. There is no middle service run by the author.

Information

updated Nov 2, 2026

Publisher
Community
Transport
stdio
Auth
None
Version
0.3.1
License
MIT
Last commit
Aug 8, 2025
Open issues
25

GitHubChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 2
  1. Web Search MCP README · github.com
  2. Web Search MCP src/index.ts · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.