What is the gateway

The Ansvar gateway is one MCP server — one URL, one OAuth flow, one tool catalogue — sitting in front of a fleet of 176 specialist MCP servers (as of 2026-07-01). The gateway is what your AI client connects to; the fleet is what answers regulatory questions.

One URL, one tool catalogue

The gateway publishes a small, curated set of orchestration tools to whichever client you connect. The surface is tier-scoped — the authoritative list is what tools/list returns for your tier. The core: search for cross-corpus lookup, get_provision for article-level retrieval, validate_citation for round-trip verification, start_workflow and its siblings for multi-stage analyses, and a handful of receipts and scoring tools for the audit path. Your client sees a stable interface that doesn't change when we add a new jurisdiction behind the gateway.

What's behind the gateway

The fleet is split by purpose (counts as of 2026-07-01, read from the routing data — never typed into this page):

55 law servers routable today, serving the 27 audited jurisdictions — primary legislation, secondary regulations, case law and preparatory works where licensing allows. Corpora for 119 jurisdictions exist further back in the build pipeline; they count as coverage only once audited and routed.
75 sector services — consolidated chassis services carrying per-country sector-regulator corpora (cybersecurity, data protection, financial supervision, competition) across the EU/EFTA states.
46 domain servers for framework material that isn't tied to a national legislator: CVE, STRIDE, OWASP, IEC 62443, and other sector-specific safety and security frameworks.

When you call search(query=..., jurisdictions=[...]) or search(query=..., sectors=[...]), the gateway picks the right MCPs, fans out, and merges the results with citations preserved at item level. See coverage for the live fleet inventory.

What the gateway is not

Not a chatbot. The gateway never generates an answer — it returns data with citations. Your AI client (Claude, Cursor, Copilot, ChatGPT) is the one synthesising prose. The gateway's job is to make sure the synthesis is grounded in a real source.
Not a RAG store. There is no embedding index inside the gateway. Each MCP behind it owns its corpus, indexes it natively (full-text search, regulator-specific schemas), and returns the precise article the agent asks for.
Not silently degrading. If an MCP is down or a jurisdiction isn't covered, the tool returns an explicit data-source-unavailable result and the client must surface it. The gateway will not fall back to a generic LLM answer when its data source fails.