Web Tools Plugin for LM Studio

An LM Studio plugin that gives local LLMs four web-oriented capabilities: searching the web, searching for images, visiting a page and reading its content, and downloading images into the chat's working directory so the assistant can embed them inline.

Key features

• Bot-resistant requests — outbound requests use impit, which applies real-browser TLS fingerprints and request headers for improved bot-detection evasion without the weight or security surface of a full browser stack.
• SSRF protection — every outbound URL is validated by dssrf before the request is issued, rejecting non-HTTP schemes, malformed URLs, and addresses that resolve to private or reserved IP ranges. A randomized double DNS resolution mitigates DNS rebinding.
• Resilient request pipeline — all requests are cached, rate-limited, and retried with randomized exponential backoff.
• Hard payload caps — separate per-image and per-page byte limits keep memory exposure bounded when fetching arbitrarily large remote content.
• Search result enrichment — search results optionally run through metascraper to surface publication date, type, and description, providing useful detail without needing to load the full page into context.
• Intelligent content extraction — Visit Website routes HTML through @mozilla/readability, stripping navigation, sidebars, and ads, so the model only sees relevant content.
• Markdown or plain-text rendering — HTML content can be converted to Markdown, or stripped to plain text for token efficiency. PDFs are transparently converted to Markdown.
• Content fuzzy matching — when a page is larger than the configured content length, relevant content is sliced and returned based on fuzzy-matched search terms, letting context-constrained models extract useful information from long pages.

Tools

Web Search

Runs a query-string search against DuckDuckGo and returns relevant results.

Parameter	Type	Notes
query	string	Required.
page	int 1–100	Optional. Defaults to 1. Enables pagination.

Returns an array of records:

Field	Type	Notes
title	string
url	string
snippet	string	Omitted when Web Search: Include Result Snippets is off.
date	string	Publication or modification date. Omitted when unavailable or enrichment is off.
type	string	OpenGraph og:type. Omitted when unavailable or enrichment is off.
description	string	Page meta description, ~500 char max. Omitted when unavailable or enrichment is off.

Image Search

Runs a query-string search against Bing's image index and returns candidate image URLs along with the source page.

Parameter	Type	Notes
query	string	Required.
page	int 1–100	Optional. Defaults to 1.

Returns an array of records:

Field	Type	Notes
image	string	Full-resolution remote URL.
title	string	Omitted when unavailable.
sourcePage	string	Page hosting the image. Omitted when unavailable.

Visit Website

Fetches a URL and returns its title, top-level headings, and a readable-content excerpt. Handles HTML pages and PDFs (and surfaces plain-text and JSON responses too). Content is returned as Markdown by default, with plain text as an option.

When a page is longer than the configured character budget, the Assistant has the option to refine the returned content by an array of search terms.

Parameter	Type	Notes
url	URL	Required.
findInPage	string[]	Optional. Biases content slices on pages exceeding the character budget.

Returns a single record:

Field	Type	Notes
url	string	Echoed input URL.
kind	string	One of html, pdf, text, or json.
mimeType	string	Server-reported or sniffed MIME type.
title	string	Omitted when unavailable.
h1	string	First <h1>. Omitted when absent or non-HTML.
h2	string	First <h2>. Omitted when absent or non-HTML.
content	string	Excerpt, truncated to the configured character budget.
contentLength	number	The length of the page content before any truncation.

Fetch Images

Downloads images into the chat's working directory and returns a Markdown reference so the assistant can embed them inline. Inputs are HTTP(S) URLs picked from Image Search, discovered in a Visit Website excerpt, or scraped automatically from a page when websiteURL is supplied.

Parameter	Type	Notes
imageURLs	URL[]	Optional. Explicit list of HTTP(S) URLs to download.
websiteURL	URL	Optional. Page to scrape for <img> tags.
maxImages	int 1–200	Optional. Overrides the scraped-image cap when websiteURL is supplied.

Returns an array of records:

Field	Type	Notes
filename	string	Derived from the URL.
alt	string	<img> alt text when scraped via websiteURL. Empty for explicit imageURLs.
title	string	<img> title attribute when scraped via websiteURL. Empty for explicit imageURLs.
image	string	Markdown image reference (). Present on success.
error	string	Failure message, in place of image on download failure.

Installation

From the LM Studio Hub

Install via the LM Studio CLI:

lms get npacker/web-tools

Or browse to the plugin on the LM Studio Hub and click "Run in LM Studio." Once enabled, the included tools become available to any model that supports tool calls.

Usage

With the plugin enabled you can explicitly ask the assistant to search the web, search for images, visit a page, or fetch images, but you can also just ask a question whose answer requires the web and the assistant will pick the right tool on its own. Image displays follow the natural pipeline: Image Search surfaces candidate URLs, Fetch Images downloads the picks into the chat's working directory, and the assistant embeds the returned Markdown references in its reply.

Configuration

All fields are exposed in the LM Studio plugin UI.

General

Field	Range	Default	Purpose
Browser Fingerprint	Firefox (latest) / Firefox 144 / Chrome (latest) / Chrome 131	Firefox (latest)	TLS and header fingerprint for outbound requests.
Safe Search	strict / moderate / off / Auto	Auto → moderate	Safe-search mode for web and image search.

Web Search

Field	Range	Default	Purpose
Web Search: Include Result Snippets	on/off	on	Include preview snippet alongside title and URL.
Web Search: Enrich Results	on/off	on	Fetch each result page to extract date, OpenGraph type, and description.
Web Search: Limit Results	on/off	on	Off includes every result on the page (≈30).
Web Search: Max Results	1–30	10	Cap when limiting is on.

Image Search

Field	Range	Default	Purpose
Image Search: Limit Results	on/off	on	Off includes every image on the page (≈35).
Image Search: Max Results	1–35	10	Cap when limiting is on. Max of 35 (Bing's page size).

Visit Website

Field	Range	Default	Purpose
Visit Website: Content Format	Markdown / Plain text	Markdown	Output format for the content field.
Visit Website: Content Character Limit	1000–100 000	10 000	Character budget for the page excerpt.
Visit Website: Max Response Size (MB)	1–100	5	Caps HTML payload (Visit Website and Fetch Images websiteURL scraping).

Fetch Images

Field	Range	Default	Purpose
Fetch Images: Max Images	1–200	10	Max images scraped from websiteURL. Overridable via maxImages.
Max Image Size (MB)	1–100	10	Caps per-image download size.

Cache TTLs

Field	Range	Default	Purpose
Search Cache TTL (s)	0–3600	900 (15 min)	Web-search payload cache lifetime. 0 disables.
Website Cache TTL (s)	0–3600	600 (10 min)	HTML cache lifetime. Shared by Visit Website, Fetch Images, and Web Search enrichment. 0 disables.

Request pacing and retry

Field	Range	Default	Purpose
Min Interval Between Requests (s)	0–30	5	Minimum gap between outbound HTTP requests. 0 disables.
Max Retries Per Request	0–4	2	Retry attempts after the first try. 0 disables.
Retry Initial Backoff (s)	0–30	1	Base backoff before the first retry.
Retry Max Backoff (s)	0–300	30	Upper bound on backoff delay.

Caching

Two disk-backed cacache stores, persisted under ~/.lmstudio/plugin-data/lms-plugin-duckduckgo-cache/:

• Web-search results — up to 100 entries, default TTL 15 minutes (searchCacheTtlSeconds). Stores the post-enrichment payload, so warm queries skip both the DuckDuckGo fetch and enrichment.
• Website HTML — up to 50 entries, default TTL 10 minutes (websiteCacheTtlSeconds). Shared by Visit Website, Fetch Images, and the Web Search enrichment pass.

Image search is not cached — Bing's response is small and the rate limiter alone caps fetch rate.

Caches survive plugin reloads. To fully clear them, stop LM Studio and delete the cache directory above.

Rate limiting and retry

A shared bottleneck-backed rate limiter enforces a minimum gap between outbound requests for single-target requests (Web Search, Image Search, Visit Website, Fetch Images). Web Search enrichment uses a per-host limiter, so requests to distinct domains run in parallel.

Every outbound request is wrapped by p-retry with randomized exponential backoff.

Local development

git clone https://github.com/packern/web-tools.git
cd web-tools
npm install
npm run dev    # runs `lms dev`

npm run push publishes the plugin to the LM Studio Hub (lms push).

Credits

Built on top of Daniel Sig's original lms-plugin-duckduckgo and lms-plugin-visit-website plugins, now merged into a single tool suite, Markdown-formatted website content, fuzzy-matching for content slicing, configurable rate limiting and retry, and a persistent cache for web-search results and fetched HTML.

License

MIT — see LICENSE.