Name: Websearch
Author: SK-DEV-AI

Websearch MCP

Multi-engine web search, fetch, browser control. 11 tools, keyless lanes.

Grade A Local · stdio No auth needed 20 tools v1.10.0 MIT

No reviews yet by SK-DEV-AIPyPI 2,132 Runs on your machine, nothing hosted

93/100

trust score

downloads a day

local

runs on your machine

166

downloads in 30d

What it does

Multi-engine web search, fetch, browser control. 11 tools, keyless lanes.

Quickstart

# 1 — add it to Claude Code
claude mcp add --scope user sk-dev-ai-websearch-mcp '--' uvx --from skdev-websearch-mcp websearch-mcp

# 2 — ask your agent for something these tools do
#   doctor, ping, search and 17 more

Websearch costs nothing on mcprush: there is no plan to choose here, no cap to set and nothing mcprush can bill you for.

Publishernot claimed

SK-DEV-AI2 servers

2 categories

All 2 servers

Collected from a public index. Nobody has claimed this account, so nothing here was written by its author — claim it if it is yours.

Installation

Where are you running it?

Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts Websearch as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. Its scan result is on the Security tab — read the surface before you approve it.

Claude Codedeclared by this listing

This is a public server: you run it yourself and this marketplace is not in the path. You need uv (docs.astral.sh/uv): uvx comes with it. Claude Code registers it in one command; --scope user makes it available in every project. The '--' is in quotes so that PowerShell passes it on.

claude mcp add --scope user sk-dev-ai-websearch-mcp '--' uvx --from skdev-websearch-mcp websearch-mcp

Reconnect, or start a new session, and the tools appear in the model’s tool list.

Free on mcprush, and it runs from its publisher’s own package: this marketplace is not in the path, so there is no key of ours to hold and no card to add.

What it asks of you

Nothing is required, but it can take GROQ_API_KEYS (Optional. Groq keys for synthesis and query rewriting.), TAVILY_KEYS (Optional. Tavily search lane.), NV_KEY (Optional. NVIDIA embeddings for ranking.), SE_KEY (Optional. Stack Exchange key (10k/day instead of 300).), S2_KEY (Optional. Semantic Scholar key (1 RPS).) if you want to set them.

What it writes

One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.

What you approve

20 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.

Tool surface

What the model actually sees. Descriptions are diffed on every release — see version history.

Only tools that write or reach the network

Read from its source code. A server that builds its tool list when it starts can list more tools than this.

Preflight: report which lanes are usable RIGHT NOW and the exact command to enable the rest. Run this first, and whenever a feature returns empty results. Distinguishes 'broken' from 'needs optional setup' — the browser lanes (Google AI Mode, browser_action, bot-wall bypass) require a…

Parameter	Type	Required	Notes
probe_network	boolean	no	Run an outbound HTTPS check (adds ~1s)

Lightweight connectivity check — verifies internet and key search endpoints are reachable. Use before expensive calls when connectivity is uncertain. No params needed. e.g. ping()

No parameter schema on file.

Multi-engine web search with dedup, reranking, and synthesis. Use for open research questions; for a known URL use fetch, for papers use arxiv, for encyclopedic facts use wikipedia. depth=1 returns snippets with relevance_score+fetch_relevance per result. depth>=2 fetches full pages + re-ranks.…

Parameter	Type	Required	Notes
query	string	yes	—
count	integer	no	—
depth	integer	no	Effort: 1=snippets only, 2+=fetch the top pages and rerank on full text. Defaults to 3 when a model lane is available (GROQ_API_KEYS or the in-tool model) and 1 otherwise; WEBSEARCH_DEFAULT_DEPTH pins
search_type	string	no	—
timelimit	string	no	Time filter: d/w/m/y
safesearch	string	no	—
language	string	no	Content language for Wikipedia/summaries
upload_urls	array	no	GAI file upload: supported formats .avif .bmp .heic .heif .jpeg .pdf .png .webp. 10MB max. Only one file per call (GAI drops all but the last). Local: file:///path or remote URL.
start_date	string	no	Tavily date filter start (YYYY-MM-DD)
end_date	string	no	Tavily date filter end (YYYY-MM-DD)
include_domains	array	no	Tavily domain include filter
exclude_domains	array	no	Tavily domain exclude filter
synthesize	boolean	no	Synthesize a cited answer from the top results. The answer is deliberately short and says so when it is truncated.
synthesize_length	string	no	short = at most 6 sentences; long = up to 12, for genuinely multi-part questions
follow_up	boolean	no	Allow ONE extra retrieval round when the first answer admits a gap (rests on a single source, a claim could not be tied to its source, or it says part of the question is unanswered). Adds 20-40s when
domain	string	no	AnySearch vertical: finance, code, academic, health, travel, legal, security. Guessed from query via simple heuristic — may be wrong, omit for general search.
anysearch_tag	string	no	AnySearch precise sub-domain tag in {domain}.{sub_domain} format (e.g. code.doc, finance.us_stock). Overrides domain.
anysearch_zone	string	no	AnySearch geo zone (cn or intl)
anysearch_language	string	no	AnySearch content language (e.g. en, zh-CN)
google_ai_only	boolean	no	Skip all other search engines and answer only via Google AI Mode. Requires a signed-in Chromium session reachable over CDP (WEBSEARCH_CDP_URL); on a headless server or without a browser this returns a

URL to markdown/text. Auto-fallback: direct fetch → CDP for blocked/JS pages. Supports PDF, EPUB, DOCX. SSRF-protected. Use focus="query" to filter content by relevance. Paginated via offset (response: next_offset). Results cached 1h; cache_ttl=0 fresh. For structured JSON extraction use extract…

Parameter	Type	Required	Notes
url	string	yes	—
max_chars	integer	no	Chars to return per call (for pagination)
offset	integer	no	Char offset for paginated reads (0 = start). Response includes is_truncated, next_offset, total_extracted_chars (full page size).
focus	string	no	BM25 relevance filter — extract only content blocks relevant to this query. Use ONLY to pull specific sections from a large page (e.g. focus='pricing'). On crucial pages INSTEAD omit focus and fetch t
cache_ttl	integer	no	Cache TTL in seconds (0 = force fresh fetch). Cache keyed by URL+extraction_type+css_selector, not focus/offset.
cache_bust	boolean	no	Ignore a fresh cache entry and re-fetch. The response carries cache with hit/age_s/ttl_remaining_s either way, so 'is this stale?' is answerable without guessing.
css_selector	string	no	CSS selector — extract only the matching element's text (CDP tier only; ignored on the direct-fetch tier)
extraction_type	string	no	—
target_language	string	no	—
output_format	string	no	—
fast	boolean	no	—
raw	boolean	no	Skip CDP/trafilatura, return raw text directly (use for GitHub raw files, pastebin, etc.)
network_idle	boolean	no	Wait for network idle before extracting (slower but captures JS-rendered content; CDP tier only)
include_images	boolean	no	Include image captions/alt text
include_links	boolean	no	Include hyperlinks in output
include_formatting	boolean	no	Preserve text formatting (bold, italic, etc)
include_tables	boolean	no	Extract tables from HTML
start_line	integer	no	1-based start line for reading a range (slices content by newline)
end_line	integer	no	1-based end line (inclusive). Use with start_line for targeted reading.
quality_floor	number	no	Min acceptable extraction quality 0-1 (length+density+structure score); below it the result is flagged low_quality

ARIA accessibility snapshot (LLM-readable page text). Use to see a page's interactive structure and what is actually visible; for readable article content use fetch. e.g. screenshot(url='https://example.com')

Parameter	Type	Required	Notes
url	string	yes	—
max_chars	integer	no	—
depth	integer	no	ARIA snapshot tree depth limit
verbose	boolean	no	Show all ARIA roles (not just interactive)
start_line	integer	no	1-based start line for snapshot text range
end_line	integer	no	1-based end line (inclusive) for snapshot text range

Drive a real browser page (needs a Chromium-based browser with --remote-debugging-port on this host, e.g. chromium/google-chrome/brave/Helium, set via WEBSEARCH_CDP_URL; run doctor() to check): navigate, click, type, press a key, evaluate JS, read page HTML/text, get the accessibility tree, or…

Parameter	Type	Required	Notes
action	string	no	navigate=go to url, click=mouse click on selector, type=insert text (focuses selector first if given), key=press a key (Enter/Escape/Tab), evaluate=run JS and return its value, html=page HTML, text=in
url	string	no	URL to act on (required for action=navigate; optional for other actions — the tab navigates there first). SSRF-validated.
selector	string	no	CSS selector — required for action=click, optional focus target for action=type, element for action=text
text	string	no	Text to type (action=type)
key	string	no	DOM key value to press (action=key), e.g. Enter, Escape, Tab, ArrowDown
expression	string	no	JavaScript expression to evaluate in the page (action=evaluate). Arrow functions are auto-invoked.
wait_state	string	no	Load milestone to wait for after navigating (and for action=wait)
steps	array	no	Ordered multi-step flow executed in ONE tab (max 25). Each step is an object with the same fields as the top-level params (action, url, selector, text, key, expression, wait_state). Use for flows like
session_id	string	no	Reuse a persistent tab from browser_session(action='create') instead of opening a throwaway one. Required for anything stateful across calls: logins, cookies, multi-page flows, forms. Without it, each
timeout	number	no	Per-step timeout in seconds (1-120)
max_chars	integer	no	Max chars returned for html/text/aria (200-200000)
full_page	boolean	no	action=screenshot: capture beyond the viewport
image_type	string	no	action=screenshot: image format
quality	integer	no	action=screenshot: JPEG quality 0-100
depth	integer	no	action=aria: accessibility tree depth (1-100)

Persistent browser tab you can reuse across calls (needs a Chromium with --remote-debugging-port; run doctor()). Without a session every browser_action call gets a fresh tab that is closed immediately, so logins, cookies and multi-step flows cannot survive. Create a session, then pass session_id…

Parameter	Type	Required	Notes
action	string	no	—
session_id	string	no	Required for action=info/close; omitted from action=close closes every session
label	string	no	Human label folded into the generated id, e.g. 'login'
url	string	no	Optional page to open in the new session

Fetch several URLs in one call. Replaces a loop of fetch calls or a throwaway requests script: dedups URLs, caps concurrency, honours robots.txt, shares the cache, and returns per-URL success/status/content plus counts. Reports truncated when the batch cap (50) is hit so you know to call again…

Parameter	Type	Required	Notes
urls	array	yes	—
max_concurrent	integer	no	Parallel fetches (1-8)
max_chars	integer	no	—
cache_ttl	integer	no	Shared cache TTL in seconds; 0 forces a refetch of every URL.
focus	string	no	Keep only content relevant to this query
respect_robots	boolean	no	—

Breadth-first crawl of a site with hard caps. Replaces a bespoke scraper and makes politeness non-optional: per-page cap, depth cap, robots.txt, dedup, concurrency limit. Returns each page's title and trimmed content plus a stop_reason so you know whether the site was exhausted or a cap stopped…

Parameter	Type	Required	Notes
start_url	string	yes	—
max_depth	integer	no	Link hops from start (0-5)
max_urls	integer	no	Hard page cap (1-100)
same_domain	boolean	no	—
include_patterns	array	no	Only URLs containing one of these substrings
exclude_patterns	array	no	—
concurrency	integer	no	—
respect_robots	boolean	no	—
max_chars	integer	no	Content kept per page

Every link on a page with its anchor text, title and rel. Replaces hand-parsing HTML before crawling or finding a 'next' button. Resolves relative hrefs, drops javascript:/mailto:, dedups, and can stay on one domain. Use before crawl or to locate a specific navigation target.

Parameter	Type	Required	Notes
url	string	yes	—
same_domain	boolean	no	—
max_links	integer	no	—
include_text	boolean	no	—

Pull HTML tables into json, csv or markdown. Replaces a pandas/lxml script every time a page has a pricing or stats table. Reports tables_found so you can pick one by index, and marks truncated when max_rows is hit. Use extract(css) for non-tabular structure instead.

Parameter	Type	Required	Notes
url	string	yes	—
selector	string	no	CSS selector to narrow which tables are considered
fmt	string	no	—
index	integer	no	0-based index among matched tables
max_rows	integer	no	—

Save a URL to disk and return the local path, byte count, sha256 and content type. Replaces curl/requests plus a filename guess. Writes only inside dest_dir (default ~/.cache/websearch/downloads) and refuses path traversal. Marks truncated if the byte cap stopped the transfer, so you know the file…

Parameter	Type	Required	Notes
url	string	yes	—
dest_dir	string	no	—
filename	string	no	—
max_bytes	integer	no	—

Has this page changed since the last time it was fetched? Compares the live page against the cached snapshot and returns changed/old_hash/new_hash plus the added and removed lines. Replaces a hand-rolled polling and hashing script. Returns had_previous=false on the first call for a URL: fetch it,…

Parameter	Type	Required	Notes
url	string	yes	—
extraction_type	string	no	—

Fill several form fields, click submit, and optionally read the result - in one call. Replaces a multi-step browser_action program full of fragile selectors. Works on a session so the submitted state is real; without session_id it creates a temporary session and closes it afterwards. fields[] is…

Parameter	Type	Required	Notes
fields	array	yes	[{selector: '#email', value: 'a@b.c'}, {selector: '#tos', action: 'click'}]
session_id	string	no	Reuse a persistent tab; omit to use a temporary one
url	string	no	Page to open first
submit_selector	string	no	Submit button; omit to fill without submitting
wait_state	string	no	—
read_selector	string	no	CSS selector to read after submitting; empty reads the whole page
timeout	number	no	—

Click a 'next' control repeatedly and collect a value from each page. Replaces a pagination loop script. Stops at max_pages, when the next control is not clickable, or when the collected value stops changing (which usually means the last page). Always returns stop_reason and marks truncated, so…

Parameter	Type	Required	Notes
next_selector	string	yes	CSS selector for the next-page control
session_id	string	no	—
url	string	no	Page to open first
max_pages	integer	no	Hard page cap (1-25)
read_selector	string	no	What to collect from each page; empty reads the whole page
wait_state	string	no	—
wait_ms	integer	no	Pause after each navigation
stop_when_unchanged	boolean	no	—
timeout	number	no	—

Full Wikipedia API client: articles, summaries, categories, links, pageviews, revisions, backlinks, geosearch, recent changes, random pages. Use for encyclopedic facts; for open research use search. e.g. wikipedia(query='Python', action='summary')

Parameter	Type	Required	Notes
action	string	no	search=find articles, summary=REST fast extract, summary_action=Action API+images, categories=list page cats, links=page links, pageviews=traffic stats
query	string	no	—
count	integer	no	—
language	string	no	—
lat	number	no	—
lon	number	no	—
category	string	no	Category name for categorymembers action (without Category: prefix)
namespace	integer	no	Namespace filter (0=articles, 14=categories). Applies to search/links/allpages/random actions.
days	integer	no	Days of pageview history (pageviews action)
type_filter	string	no	Change type filter for recentchanges: edit/new/move/log/categorize
include_images	boolean	no	Include thumbnail images in summary_action
distance	integer	no	Search radius in meters for geosearch

Search arXiv academic papers (single-source; for multi-source paper search use codesearch papers). Use raw_query for boolean operators (AND, OR, ANDNOT), phrase search (ti:"exact phrase"), wildcards (au:smith*). e.g. arxiv(query='cs.AI transformer', count=5)

Parameter	Type	Required	Notes
query	string	yes	—
count	integer	no	—
search_field	string	no	—
sort_by	string	no	—
sort_order	string	no	—
start	integer	no	—
id_list	string	no	Comma-delimited arXiv IDs
category	string	no	arXiv category filter (e.g. cs.AI, math.CO)
raw_query	string	no	Raw arXiv search_query syntax with boolean operators (AND/OR/ANDNOT), phrase, wildcards. Overrides query+search_field.

Discover all pages on a website via sitemap XML (primary) and HTML link extraction (fallback). Use to survey a site's pages before fetching individually. Returns the base domain, source type, total count, and an array of discovered URLs with metadata (last_modified, priority, changefreq when…

Parameter	Type	Required	Notes
url	string	yes	Full URL of the site to map (e.g. https://example.com)
max_urls	integer	no	Cap on returned URLs
max_depth	integer	no	Recursive link extraction depth (0=homepage only). Only used when no sitemap exists.
include_sitemap	boolean	no	Try sitemap discovery first
include_links	boolean	no	Fall back to HTML link extraction when no sitemap
same_domain	boolean	no	Only include URLs from the same domain
exclude_patterns	array	no	Regex patterns to exclude matching URLs
rank	boolean	no	Score each URL by fetch quality + sitemap priority and return important-first with rank_score (costs one scored pass; use before deep-fetching a site)

Extract structured JSON from a webpage using LLM, CSS, or regex strategies. Persistent cache — repeat calls free. For LLM strategy, describe what you want and get clean JSON back. For CSS strategy, provide field names (fastest, free). For raw page content, use fetch instead. Examples:…

Parameter	Type	Required	Notes
url	string	yes	Target URL to extract data from
instruction	string	no	Natural language extraction instruction (used with strategy=llm). Example: 'extract all product names, prices, and ratings from this page'
strategy	string	no	css=CSS-selector-based extraction (fast, free), llm=AI-powered (costs quota), regex=pattern-based (emails, phones, URLs)
fields	array	no	List of field names for extraction (e.g. ['name', 'price', 'rating']). Used with strategy=css or strategy=llm
chunk_threshold	integer	no	Max tokens per chunk for LLM extraction (lower = cheaper, higher = more context)
css_selector	string	no	CSS selector for the container element (used with strategy=css). Defaults to 'body'
provider	string	no	LLM provider string in LiteLLM format (e.g. groq/openai/gpt-oss-120b, openai/gpt-4o, ollama/llama2)

PDF extraction: PyMuPDF fast text-layer path, Docling+OCR fallback for scanned/image-only PDFs (CPU). Formats: markdown (default), json, html, or comma combos like 'markdown,json'. e.g. pdf_extract(input_path='/path/to/doc.pdf', format='markdown')

Parameter	Type	Required	Notes
input_path	array	yes	PDF file paths or URLs (local files, http/https, file://)
format	string	no	—
password	string	no	PDF password for protected files
pages	string	no	Page range e.g. 1-5,8,10-12
hybrid	string	no	Force the Docling OCR fallback regardless of text layer (upper bound: CPU cost only)
hybrid_mode	string	no	Alias for hybrid (same values)
force_ocr	boolean	no	Shortcut: force the Docling OCR fallback even when a text layer exists

Free

$0 forever

Every tool, no call limit
No card, and mcprush charges nothing for it
Source published under MIT
Runs on your machine — nothing of it reaches our gateway
Nothing to cap, because mcprush bills nothing

Nothing here is billable: mcprush charges nothing to install Websearch or to call it, at any volume.

ATrust score 93 / 100

Release 1.10.0 · scanned 28 Oct 2026

Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.

Threat score100

Client score93

Findings3 threat · 0 capability

Capabilityhigh

Coveragesource

Verificationrepo

Surface digestdac4574cab6aea3fa458d039f6710053a8f053b6143c4d6eff2bea0e9e2f3a48

Score vectorthreat 100 → client 93

Term	Level	What it prices	Points
capability-exposure	high	capability blast radius (high) — client exposure if the model is manipulated	−6
verification-discount	repo	publisher verification (public source) — no provenance, but the source is public and inspectable	−1
coverage-honesty	source	inspection depth (source) — how much of the target the scan could see	−0

score = clamp(round(threatScore − E_cap − E_ver − E_cov), 0, 100). No model, no opinion, no network call at scoring time: the same methodology version against the same bytes always produces this number, and every line above is a field of the report rather than a summary of it.

What the scan could actually read

A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.

Tool surfaceread

Implementation sourceread

Package metadataread

Live transportnot available

Tools were statically extracted from the published source (18 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command "npx -y <package>".

Capability — what it could do if the model were manipulated

Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. high is the level these add up to.

Tool	Capability tags	Why the tag was assigned
doctor	no tags
ping	no tags
search	untrusted-input	untrusted-input: keyword "web_search"
fetch	untrusted-input	untrusted-input: keyword "fetch"
screenshot	untrusted-input	untrusted-input: keyword "fetch"
browser_action	no tags
browser_session	no tags
fetch_many	untrusted-input	untrusted-input: keyword "fetch"
crawl	untrusted-input	untrusted-input: keyword "crawl"
extract_links	untrusted-input	untrusted-input: keyword "crawl"
extract_table	no tags
download	untrusted-input	untrusted-input: keyword "download"
diff_page	untrusted-input	untrusted-input: keyword "fetch"
wikipedia	no tags
arxiv	no tags
map_site	no tags
extract	no tags
pdf_extract	no tags

Toxic-flow graph

The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.

The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.

Supply chain and provenance

Integrityfirst-seen

Surface digestdac4574cab6aea3f

Scanned2026-10-28

This is the first scan of this surface here, so there is nothing yet to compare it against.

Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.