Search peer-reviewed papers across arXiv, PubMed, IEEE, Nature, Springer, and 12+ scholarly databases via site-restricted web search. Returns JSON with fields: papers (array of {title, url, source, abstract}), query, totalResults, resultCount, source. No special query syntax needed — use technical terms directly. year_from/year_to combine with source filter to narrow scope. On no matches returns resultCount: 0 with empty array; on failure returns isError with message. Subject to per-tenant rate limit (default 30 req/min) with automatic provider fallback. Use this for scientific research, literature reviews, or citations — not for general content. Use web_search for non-academic technical content, or news_search for recent scientific announcements. Set pdf_only=true when you plan to pass URLs to scrape_page for full-text extraction. Results cached 24 hours.

Find images by query with filters for size, type, color, and format. Returns JSON with fields: images (array of {title, link, thumbnailLink, displayLink, contextLink, width, height, fileSize}), query, resultCount. Results sorted by relevance; max 10 per call, no pagination. On no matches returns resultCount: 0 with empty array; on failure returns isError with message. Subject to per-tenant rate limit (default 30 req/min) with automatic provider fallback. Use this for visual assets or image references — not for text information. Use web_search for pages containing images, or scrape_page to extract images from a known URL. Results cached 30 min.

Search recent news articles by topic with time-based freshness filtering. Returns JSON with fields: articles (array of {title, url, source, publishedAt, snippet}), query, resultCount. Default freshness is 'week'; use 'hour' or 'day' for breaking news. On no matches returns resultCount: 0 with empty array; on failure returns isError with message. Subject to per-tenant rate limit (default 30 req/min) with automatic provider fallback. Coverage depends on configured search provider. Use web_search instead for evergreen/non-news content; use academic_search for peer-reviewed findings; use search_and_scrape if you need full article text beyond snippets. Results cached 15 min due to news volatility.

Search Google Patents for prior art, competitive landscapes, or specific patents by number. Returns JSON with fields: patents (array of {title, url, number, abstract}), query, searchType, resultCount. Query accepts patent numbers (e.g. 'US11234567') or natural-language invention descriptions. search_type adjusts strategy: prior_art (broad technical), specific (exact lookup), landscape (competitive overview). Auto-generates assignee name variations (with/without Inc, LLC, Corp, Ltd suffixes). On no matches returns resultCount: 0 with empty array; on failure returns isError with message. Subject to per-tenant rate limit with provider fallback. Use academic_search for published research papers, or web_search for general technical content. Results cached 24 hours.

Extract readable content from a single URL. Handles web pages (HTML and JS-rendered SPAs), PDFs, DOCX, PPTX, and YouTube transcripts via auto-detected extraction with tiered fallback (markdown, stealth, HTML, headless browser). Returns JSON with fields: url, content, contentType, contentLength, truncated, estimatedTokens, sizeCategory, citation (with formatted APA/MLA), metadata ({title, author}). Content capped at max_length (default 50000 bytes); truncated=true if cut. Mode 'preview' forces max_length to 5000 bytes for quick relevance checks. Max 5 concurrent scrapes; additional calls queue. On failure returns isError with reason (e.g. blocked, timeout, invalid URL). Use search_and_scrape instead to discover and extract in one step; use web_search if you only need URLs. Results cached 1 hour.

Search the web and extract full content from top results in one call. Scrapes in parallel (max 5 concurrent), deduplicates content across sources, and scores each source on relevance and quality. Returns JSON with fields: query, combinedContent, sources (array of {url, title, content, contentType, scores} — included when include_sources=true), summary ({urlsSearched, urlsScraped, processingTimeMs}), sizeMetadata ({totalLength, estimatedTokens, sizeCategory}). On zero search matches returns empty combinedContent with urlsSearched: 0. Individual scrape failures are silently skipped (urlsScraped < urlsSearched indicates partial failures). num_results controls sources scraped (more = slower, typically 2-15s). Subject to per-tenant rate limit with provider fallback. Use web_search instead if you only need URLs; use scrape_page for a single known URL. Not cached (combines live search + scrape).

Track multi-step research progress with persistent sessions, branching, and knowledge gap tracking. This is a state tracker, not a search tool — pair with web_search or search_and_scrape for actual searches. Call with stepNumber=1 (omit sessionId) to start a new session. Returns JSON with fields: sessionId, currentStep, totalStepsEstimate, isComplete, steps (array of {stepNumber, description, isRevision, revisesStep, branchId, timestamp}), sources, gaps (array of {description, foundInStep}), startedAt, completedAt (only when isComplete=true). Pass sessionId for steps 2+; each call returns the full accumulated session state. Set nextStepNeeded=false on final step to mark complete. Sessions expire after 30 min of inactivity (returns 'session expired or not found' error); max 50 concurrent sessions per tenant. Use branchFromStep + branchId to explore alternative directions without losing the main thread. Not cached.

Search the web for URLs and metadata without fetching page content. Returns JSON with fields: urls (string array), results (array of {title, url, snippet, displayLink}), query, resultCount. lens and site are mutually exclusive (lens overrides site if both provided). On no matches returns resultCount: 0 with empty results array; on failure returns isError with message. Subject to per-tenant rate limit (default 30 req/min) with automatic provider fallback and circuit breaker recovery. Supports lenses (programming, news, tech, legal, medical, finance, science, government) for domain-restricted search. Use search_and_scrape instead when you need full page content; use news_search for time-sensitive current events; use academic_search for scholarly papers. Results cached 30 min; use time_range to constrain freshness when cache staleness is a concern.