Most of these stop at markdown. Here a screenshot or a screencast becomes a baseline the next run is diffed against — and arachne_visual_gates checks overflow, text collision and JS errors with no baseline at all. That is a deploy check wearing a scraper's clothes.
A broad web and media toolkit backed by the Arachne API. Search and scrape, drive a real browser past bot detection, run vision over images, PDFs, screenshots and video, transcribe audio, query a RAG knowledge base, and diff a page against a visual baseline.
- Web and data — arachne_search, arachne_scrape for clean markdown, arachne_extract as the catch-all for YouTube, audio, video, PDF, DOCX and images, arachne_browser_extract with evasion for sites that block scrapers, arachne_browser_run for scripted clicks and logins, arachne_query against the knowledge base
- Capture — arachne_screenshot, arachne_record for a .webm screencast, and arachne_screenshot_vision and arachne_record_vision to run the vision pipeline over what was captured
- Vision and audio — arachne_vision for OCR, colours, quality, faces and edges, arachne_pdf_vision, arachne_analyze_video, arachne_transcribe through Whisper
- Visual regression — arachne_visual_snapshot creates or compares a baseline, arachne_visual_diff returns a diff ratio, arachne_visual_gates runs key-free checks for overflow, text collision and JS errors with no baseline needed, arachne_visual_report and arachne_visual_video_report build side-by-side HTML, arachne_visual_approve promotes a capture to baseline, plus arachne_visual_list, arachne_visual_video and arachne_visual_failure_bundle
- Code knowledge graph — code_graph, kg_context for a ranked context extract, kg_communities, kg_god_nodes by PageRank, kg_watch for incremental change detection over git diff
- Packages and repos — arachne_download, arachne_repo_download, arachne_repo_fork, arachne_repo_forks, arachne_repo_index, arachne_releases, arachne_release_download, arachne_pypi_download, arachne_npm_download, arachne_gitlab_download
- GitHub research — arachne_github_trending, arachne_github_user_prs, arachne_github_user_activity, arachne_github_repo_metadata
- Desktop and agentic — arachne_desktop, arachne_desktop_os, arachne_observe with numbered clickable bounding boxes, arachne_screen_analyze
- Utilities — arachne_plan, arachne_capabilities, arachne_format_converter, arachne_calc, arachne_metrics, arachne_sandbox_status, arachne_ping, arachne_social_research, arachne_youtube_comments, arachne_torrent_extract
The arachne-mcp package from PyPI over stdio, with ARACHNE_API_KEY in the environment from the Arachne developer page. ARACHNE_BASE_URL points at another instance and defaults to https://arachne.seu.pet. Everything runs against the hosted API, so the browser, vision and transcription work needs no Playwright or Whisper locally. Desktop control targets X11 or WSLg. Many tool descriptions are written in Portuguese.
One command plus a key — pip install arachne-mcp, then supply credentials