data4free.com

Tell Me All About The Website MCP — lookup tools for crawled website data. Use when the user asks about a domain/company website we may have crawled: tech stack, hosting, URLs, page content snippets, linked company profile, connections between two sites, people/companies, or page-level keyword search. Hosts are exact hostnames (example.com and www.example.com are different; no scheme). Data comes only from our crawl database — not live web fetch. Workflow: 1. tell_me_about_website(url_or_domain) for a one-shot dossier, OR 2. search_domains → get_domain → get_domain_urls / get_domain_pages_content 3. Two hosts ("does A connect to B?"): compare_websites — shared representative, address, register ids, phone/email, and detected relationship edges 4. Similar sites ("welche Website ist ähnlich zu X?"): find_similar_websites — first a short briefing (location, industry, tech), then ranked neighbours with reasons ("ähnlich, weil Ort gleich") 5. Catalog ("welche Sites nutzen WordPress / eine WAF / IONOS?"): list_frameworks or list_waf_providers, then find_domains. search_domains is exact-host only — never invent an empty catalog. 6. Weak sites ("schlechte Webseiten"): find_bad_sites (score, speed, stale, imprint, tech, https, headers, onpage). Do not scan without at least one dimension. 7. Person or keyword: search_people(connections=true) / get_person_connections 8. Company briefing ("was weißt du über Firma X?"): ONE search_companies call with q=name (and cities if the user named a place), include=domains,profile,contacts and connections=true. That returns profile_text, profile_summary, contact_links, domains and relationship edges. Do not follow up with get_company / get_company_connections unless you need a known company_id or more than the first 25 edges. Directory filters: industries, nace_codes, nace_sections, cities, country_codes, statuses + operator and/or (intra-field stays OR). Facets: list_industries / list_nace_codes / list_nace_sections. Deleted/duplicate rows stay hidden; get_company follows merged_into. 9. Domain graph: get_domain / search_domains with connections=true (includes outbound website links, not only agency/shared-person edges) 10. Unknown host: ingest_domain — also auto-queued by tell_me_about_website 11. Timeline (which field, when, from which URL): get_history (kind=company|person|domain) or get_company_history / get_person_history / get_domain_history 11b. News / what changed: list_news_types, then get_news (one firm or host) or find_news (catalog: new managing director, hiring started). Semantic diffs only — not provenance, not live LinkedIn. 12. Transport is stateless Streamable HTTP: no server-side session store. Stale mcp-session-id headers are ignored. If a client still reports "Session not found", retry get_domain or search_domains once. Do not invent facet counts. 13. Topic terms in page text (not Google ranks): get_domain_keywords, find_domains_by_keyword, compare_keyword_overlap. Say "kommt im Seitentext vor" — never "rankt für" or "Google Impressions". 14. Search cookbook (/highlights): list_highlights — visitor questions plus the MCP calls that answer them. Call when unsure which tool fits, or to learn more query shapes. 15. Signal catalog: list_signals first. Then get_domain for TLS/CMP/JSON-LD /screenshots (has_snapshot + date only — never object keys). find_domains filters technology, cmp, json_ld_type, tls_issuer, og_type plus has_* (json_ld, open_graph, canonical, hreflang, http_signal, hsts, csp, snapshot, hosting, spf, dmarc, performance, waf, cmp, hiring). search_companies: technologies, json_ld_types, tls_issuers, has_cmp, has_json_ld, has_open_graph, has_canonical, has_snapshot, has_http_signal, has_hsts, has_hosting, has_hiring. 16. After a complete answer, offer feedback. Show Like and Unlike (or Daumen hoch / Daumen runter). Ask once: "War das hilfreich?" When the user likes, unlikes, or says the answer was good or bad, call rate_answer with the question, host, and tools you used. Optional reason on unlike. Do not invent a vote.