README.md
README.md
Online web search for local LLMs. Ships in two forms from a single codebase:
mcp.json (LM Studio, Claude Desktop, etc.).Bundles 14 search engines and a parallel multi-engine mode that fans out one query, interleaves results round-robin, and dedupes by URL. Also exposes a dependency-free page reader so the model can fetch and read result URLs.
No browser, no Playwright, no API keys — just Node 22's built-in fetch and a small HTML→text extractor.
| Tool | What it does |
|---|---|
list_search_engines | Returns available engine ids and the current mode/defaults. |
web_search | Runs a search, filters junk (empty titles, redirect wrappers), dedupes by normalized URL, ranks by host trust + title/snippet query-overlap, returns the top K with a score (0–1) and host per result. |
fetch_page | Fetches one URL and returns its extracted main text. If you pass query, long pages are truncated query-aware — paragraphs containing query terms are kept instead of just the first N chars. |
web_search_full | One-stop: web_search + concurrently fetches the top N result pages with query-aware extraction. Use this instead of web_search + multiple calls. |
All search responses include returned, dropped_as_noise, from_cache, and per_engine_counts for transparency. When the top result score is very low (< 0.35) or zero results survive filtering, an additional note field tells the model to rephrase or pick a more specific tool.
Every web_search / web_search_full response is post-processed:
Turn on includeScoreBreakdown (per-chat setting / MULTI_SEARCH_SCORE_BREAKDOWN=1) to see per-result { host_trust, title_overlap, snippet_overlap, snippet_length_bonus }.
A tiny LRU+TTL cache (100 search entries, 50 web_search_full entries) deduplicates repeat calls within the same plugin session. Default TTL is 5 minutes. Set cacheTtlSec to 0 (or MULTI_SEARCH_CACHE_TTL=0) to disable. Responses include from_cache: true when served from cache.
duckduckgo, brave, bing, qwant, ecosia, startpage, metager,
wikipedia, arxiv, reddit, stackoverflow, github, google-scholar,
devdocs
Most reliable:
duckduckgo,brave,bing,wikipedia,arxiv,stackoverflow,github. Scraped engines (ecosia,startpage, etc.) can occasionally be rate-limited — the multi-engine mode tolerates this and just keeps whatever returned successfully.
Requires Node 22+ (uses the built-in global fetch / AbortController).
After npm run build, point your MCP host at dist/mcp.js. Example mcp.json (LM Studio, Claude Desktop, etc.):
MCP has no settings UI, so the standalone server is configured purely via env vars:
| Variable | Default | Range | Purpose |
|---|---|---|---|
MULTI_SEARCH_MODE | multi | single | multi | Search strategy. |
MULTI_SEARCH_SINGLE_ENGINE | duckduckgo | any engine id | Engine used in single mode. |
MULTI_SEARCH_MULTI_ENGINES | duckduckgo,brave,bing,wikipedia | comma list | Engines queried in multi mode. |
MULTI_SEARCH_MAX_RESULTS | 5 | 1–20 | Raw results per engine cap. |
MULTI_SEARCH_TOP_K | 8 | 1–25 | Final top-K returned after ranking. |
MULTI_SEARCH_SNIPPET_MAX | 240 | 80–2000 | Snippet length cap (chars). |
MULTI_SEARCH_CACHE_TTL | 300 | 0–3600 sec | Search-cache TTL. 0 = disabled. |
MULTI_SEARCH_SCORE_BREAKDOWN | 0 |
Should print two JSON-RPC responses on stdout and multi-search MCP server ready on stdio on stderr.
When loaded as a plugin you get a per-chat config panel with these fields:
| Field | Default | Purpose |
|---|---|---|
searchEnabled | ✅ on | Kill-switch. When off, no tools are exposed to the model at all. |
mode | single | single or multi. Only one web_search tool is exposed at a time, matching this mode. |
singleEngine | duckduckgo | Engine used in single mode. |
maxResultsPerEngine | 5 | Raw results-per-engine cap (1–20). |
topK | 8 | Final number of best-scoring results returned after ranking (1–25). |
snippetMaxChars | 240 | Snippet length cap to save context (80–2000). |
cacheTtlSec | 300 | Search-cache TTL in seconds. 0 disables caching. |
includeScoreBreakdown | off | Attach per-result scoring breakdown for transparency / debugging. |
fetchTopN |
And global plugin settings (apply across all chats):
| Field | Default | Purpose |
|---|---|---|
requestTimeoutMs | 15000 | Per-engine HTTP timeout. |
pageFetchTimeoutMs | 10000 | Timeout for fetch_page / web_search_full page loads. |
maxContentLength | 10000 | Hard cap on extracted page text length. |
userAgent | Chrome 120 UA | User-Agent header. |
Promise.all).?query and #fragment).source engine so the model knows where it came from.This gives broader topical coverage than a single engine and degrades gracefully when one engine breaks.
fetch_page and web_search_full use a small dependency-free extractor that:
It is not a full Readability re-implementation — for pathological pages (heavy SPA, lazy-loaded content, Cloudflare challenges) it will return less than a browser-based scraper would. The trade-off is zero native deps and instant cold start.
Both distributions share engines.ts and extract.ts, so any new engine becomes available in both modes at once.
| Command | Purpose |
|---|---|
npm run build | Compile TypeScript to dist/. |
npm run dev | lms dev — run as a hot-reloading LM Studio plugin. |
npm run push | lms push — install the plugin into LM Studio. |
npm run start:mcp | node dist/mcp.js — run the MCP server on stdio. |
MIT
Online web search for local LLMs. Ships in two forms from a single codebase:
mcp.json (LM Studio, Claude Desktop, etc.).Bundles 14 search engines and a parallel multi-engine mode that fans out one query, interleaves results round-robin, and dedupes by URL. Also exposes a dependency-free page reader so the model can fetch and read result URLs.
No browser, no Playwright, no API keys — just Node 22's built-in fetch and a small HTML→text extractor.
| Tool | What it does |
|---|---|
list_search_engines | Returns available engine ids and the current mode/defaults. |
web_search | Runs a search, filters junk (empty titles, redirect wrappers), dedupes by normalized URL, ranks by host trust + title/snippet query-overlap, returns the top K with a score (0–1) and host per result. |
fetch_page | Fetches one URL and returns its extracted main text. If you pass query, long pages are truncated query-aware — paragraphs containing query terms are kept instead of just the first N chars. |
web_search_full | One-stop: web_search + concurrently fetches the top N result pages with query-aware extraction. Use this instead of web_search + multiple calls. |
All search responses include returned, dropped_as_noise, from_cache, and per_engine_counts for transparency. When the top result score is very low (< 0.35) or zero results survive filtering, an additional note field tells the model to rephrase or pick a more specific tool.
Every web_search / web_search_full response is post-processed:
Turn on includeScoreBreakdown (per-chat setting / MULTI_SEARCH_SCORE_BREAKDOWN=1) to see per-result { host_trust, title_overlap, snippet_overlap, snippet_length_bonus }.
A tiny LRU+TTL cache (100 search entries, 50 web_search_full entries) deduplicates repeat calls within the same plugin session. Default TTL is 5 minutes. Set cacheTtlSec to 0 (or MULTI_SEARCH_CACHE_TTL=0) to disable. Responses include from_cache: true when served from cache.
duckduckgo, brave, bing, qwant, ecosia, startpage, metager,
wikipedia, arxiv, reddit, stackoverflow, github, google-scholar,
devdocs
Most reliable:
duckduckgo,brave,bing,wikipedia,arxiv,stackoverflow,github. Scraped engines (ecosia,startpage, etc.) can occasionally be rate-limited — the multi-engine mode tolerates this and just keeps whatever returned successfully.
Requires Node 22+ (uses the built-in global fetch / AbortController).
After npm run build, point your MCP host at dist/mcp.js. Example mcp.json (LM Studio, Claude Desktop, etc.):
MCP has no settings UI, so the standalone server is configured purely via env vars:
| Variable | Default | Range | Purpose |
|---|---|---|---|
MULTI_SEARCH_MODE | multi | single | multi | Search strategy. |
MULTI_SEARCH_SINGLE_ENGINE | duckduckgo | any engine id | Engine used in single mode. |
MULTI_SEARCH_MULTI_ENGINES | duckduckgo,brave,bing,wikipedia | comma list | Engines queried in multi mode. |
MULTI_SEARCH_MAX_RESULTS | 5 | 1–20 | Raw results per engine cap. |
MULTI_SEARCH_TOP_K | 8 | 1–25 | Final top-K returned after ranking. |
MULTI_SEARCH_SNIPPET_MAX | 240 | 80–2000 | Snippet length cap (chars). |
MULTI_SEARCH_CACHE_TTL | 300 | 0–3600 sec | Search-cache TTL. 0 = disabled. |
MULTI_SEARCH_SCORE_BREAKDOWN | 0 |
Should print two JSON-RPC responses on stdout and multi-search MCP server ready on stdio on stderr.
When loaded as a plugin you get a per-chat config panel with these fields:
| Field | Default | Purpose |
|---|---|---|
searchEnabled | ✅ on | Kill-switch. When off, no tools are exposed to the model at all. |
mode | single | single or multi. Only one web_search tool is exposed at a time, matching this mode. |
singleEngine | duckduckgo | Engine used in single mode. |
maxResultsPerEngine | 5 | Raw results-per-engine cap (1–20). |
topK | 8 | Final number of best-scoring results returned after ranking (1–25). |
snippetMaxChars | 240 | Snippet length cap to save context (80–2000). |
cacheTtlSec | 300 | Search-cache TTL in seconds. 0 disables caching. |
includeScoreBreakdown | off | Attach per-result scoring breakdown for transparency / debugging. |
fetchTopN |
And global plugin settings (apply across all chats):
| Field | Default | Purpose |
|---|---|---|
requestTimeoutMs | 15000 | Per-engine HTTP timeout. |
pageFetchTimeoutMs | 10000 | Timeout for fetch_page / web_search_full page loads. |
maxContentLength | 10000 | Hard cap on extracted page text length. |
userAgent | Chrome 120 UA | User-Agent header. |
Promise.all).?query and #fragment).source engine so the model knows where it came from.This gives broader topical coverage than a single engine and degrades gracefully when one engine breaks.
fetch_page and web_search_full use a small dependency-free extractor that:
It is not a full Readability re-implementation — for pathological pages (heavy SPA, lazy-loaded content, Cloudflare challenges) it will return less than a browser-based scraper would. The trade-off is zero native deps and instant cold start.
Both distributions share engines.ts and extract.ts, so any new engine becomes available in both modes at once.
| Command | Purpose |
|---|---|
npm run build | Compile TypeScript to dist/. |
npm run dev | lms dev — run as a hot-reloading LM Studio plugin. |
npm run push | lms push — install the plugin into LM Studio. |
npm run start:mcp | node dist/mcp.js — run the MCP server on stdio. |
MIT
fetch_pageget_weather | Current conditions + 3-day forecast for any city / ZIP / airport code. Backed by wttr.in (no API key). Use this for weather questions instead of web_search — search snippets almost never contain actual temperatures. |
duckduckgo.com/l/?, google.com/url?, etc.), non-http URLs.www., strip #fragment, drop utm_* / fbclid / ref / source / gclid tracking params, strip trailing /.0.45 · titleQueryOverlap + 0.30 · hostTrust + 0.20 · snippetQueryOverlap + 0.05 · snippetLengthBonus.
topK (default 8).snippetMaxChars (default 240).0/1 |
| Include per-result scoring breakdown for transparency. |
MULTI_SEARCH_FETCH_TOP_N | 3 | 0–10 | Pages fetched by web_search_full. |
MULTI_SEARCH_FETCH_CONCURRENCY | 3 | 1–8 | Parallel page fetches. |
MULTI_SEARCH_REQUEST_TIMEOUT | 15000 | 1000–60000 ms | Per-engine HTTP timeout. |
MULTI_SEARCH_PAGE_TIMEOUT | 10000 | 1000–60000 ms | Page-fetch timeout. |
MULTI_SEARCH_MAX_CONTENT | 10000 | 500–200000 chars | Extracted page text cap. |
MULTI_SEARCH_USER_AGENT | Chrome 120 UA | string | User-Agent for all HTTP requests. |
3 |
Pages fetched by web_search_full (0–10). |
fetchConcurrency | 3 | Parallel page fetches (1–8). |
Multi: <Engine> × 14 | duckduckgo / brave / bing / wikipedia | One checkbox per engine — pick which to include in multi mode. |
[]<script>, <style>, <noscript>, <svg>, <iframe>, <form>, <nav>, <aside>, <footer>, <header>,<article> → <main> → largest <div> by visible text,<p>foo</p><p>bar</p> doesn't collapse to foobar),maxContentLength.EngineFn in src/engines.ts:
ENGINES map.ENGINE_LABELS in src/configSchematics.ts.DEFAULT_MULTI if you want it ticked by default.npm run build. It now appears in both the plugin's checkbox list and the MCP server's list_search_engines output.git clone https://github.com/potkolainen/LM-Multi-Online-Search-plugin.git
cd "Multi search"
npm install
npm run build
{
"mcpServers": {
"multi-search": {
"command": "node",
"args": ["/absolute/path/to/Multi search/dist/mcp.js"],
"cwd": "/absolute/path/to/Multi search",
"env": {
"MULTI_SEARCH_MODE": "multi",
"MULTI_SEARCH_MULTI_ENGINES": "duckduckgo,brave,bing,wikipedia",
"MULTI_SEARCH_FETCH_TOP_N": "3",
"MULTI_SEARCH_MAX_CONTENT": "10000"
}
}
}
}
printf '%s\n%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"x","version":"0"}}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
| node dist/mcp.js
npm install
lms dev # hot-reload while developing
# or
lms push # install into LM Studio
src/
engines.ts # 14 search engines + shared helpers (fetch with timeout, etc.)
extract.ts # HTML → main-text extractor (with query-aware truncation)
ranking.ts # URL normalization, dedupe, host-trust scoring, top-K filter
cache.ts # Tiny in-process LRU + TTL cache
weather.ts # wttr.in client for the get_weather tool
configSchematics.ts # LM Studio plugin per-chat + global config schema
toolsProvider.ts # LM Studio plugin tool registration
index.ts # LM Studio plugin entry point
mcp.ts # Standalone MCP stdio server entry point
fetch_pageget_weather | Current conditions + 3-day forecast for any city / ZIP / airport code. Backed by wttr.in (no API key). Use this for weather questions instead of web_search — search snippets almost never contain actual temperatures. |
duckduckgo.com/l/?, google.com/url?, etc.), non-http URLs.www., strip #fragment, drop utm_* / fbclid / ref / source / gclid tracking params, strip trailing /.0.45 · titleQueryOverlap + 0.30 · hostTrust + 0.20 · snippetQueryOverlap + 0.05 · snippetLengthBonus.
topK (default 8).snippetMaxChars (default 240).0/1 |
| Include per-result scoring breakdown for transparency. |
MULTI_SEARCH_FETCH_TOP_N | 3 | 0–10 | Pages fetched by web_search_full. |
MULTI_SEARCH_FETCH_CONCURRENCY | 3 | 1–8 | Parallel page fetches. |
MULTI_SEARCH_REQUEST_TIMEOUT | 15000 | 1000–60000 ms | Per-engine HTTP timeout. |
MULTI_SEARCH_PAGE_TIMEOUT | 10000 | 1000–60000 ms | Page-fetch timeout. |
MULTI_SEARCH_MAX_CONTENT | 10000 | 500–200000 chars | Extracted page text cap. |
MULTI_SEARCH_USER_AGENT | Chrome 120 UA | string | User-Agent for all HTTP requests. |
3 |
Pages fetched by web_search_full (0–10). |
fetchConcurrency | 3 | Parallel page fetches (1–8). |
Multi: <Engine> × 14 | duckduckgo / brave / bing / wikipedia | One checkbox per engine — pick which to include in multi mode. |
[]<script>, <style>, <noscript>, <svg>, <iframe>, <form>, <nav>, <aside>, <footer>, <header>,<article> → <main> → largest <div> by visible text,<p>foo</p><p>bar</p> doesn't collapse to foobar),maxContentLength.EngineFn in src/engines.ts:
ENGINES map.ENGINE_LABELS in src/configSchematics.ts.DEFAULT_MULTI if you want it ticked by default.npm run build. It now appears in both the plugin's checkbox list and the MCP server's list_search_engines output.git clone https://github.com/potkolainen/LM-Multi-Online-Search-plugin.git
cd "Multi search"
npm install
npm run build
{
"mcpServers": {
"multi-search": {
"command": "node",
"args": ["/absolute/path/to/Multi search/dist/mcp.js"],
"cwd": "/absolute/path/to/Multi search",
"env": {
"MULTI_SEARCH_MODE": "multi",
"MULTI_SEARCH_MULTI_ENGINES": "duckduckgo,brave,bing,wikipedia",
"MULTI_SEARCH_FETCH_TOP_N": "3",
"MULTI_SEARCH_MAX_CONTENT": "10000"
}
}
}
}
printf '%s\n%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"x","version":"0"}}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
| node dist/mcp.js
npm install
lms dev # hot-reload while developing
# or
lms push # install into LM Studio
src/
engines.ts # 14 search engines + shared helpers (fetch with timeout, etc.)
extract.ts # HTML → main-text extractor (with query-aware truncation)
ranking.ts # URL normalization, dedupe, host-trust scoring, top-K filter
cache.ts # Tiny in-process LRU + TTL cache
weather.ts # wttr.in client for the get_weather tool
configSchematics.ts # LM Studio plugin per-chat + global config schema
toolsProvider.ts # LM Studio plugin tool registration
index.ts # LM Studio plugin entry point
mcp.ts # Standalone MCP stdio server entry point
async function myEngine(query: string, opts: EngineOptions): Promise<SearchResult[]> { … }
async function myEngine(query: string, opts: EngineOptions): Promise<SearchResult[]> { … }