Find public contracts worth winning. Let your AI do the searching.

It is built machine-first: the API and the MCP server are the product rather than an add-on, and the website runs on the same records. Same field names everywhere, provenance and the source's licence on every record, and limits published on the pricing page.

One URL and your API key — or an OAuth sign-in you approve once in the browser, where the assistant supports it (Claude and ChatGPT do). In Claude Desktop or claude.ai: Settings → Connectors → add the MCP endpoint. In Claude Code it's a single claude mcp add command — and ChatGPT, Perplexity and Grok each have their own one-paste path. Your agent gets all fifteen tools: search, full records, bid/no-bid qualification, award analytics and cross-session monitors, plus guided prompts. Everything is also plain REST with an OpenAPI spec.

Two tiers, and the map on this page colours them differently because they are different promises. For countries whose own national platform we ingest, you get that market to whatever depth the publisher offers — often the full below-threshold layer, and in several cases decades of history. For the rest of the EU and EEA we hold TED, which carries every member state's above-threshold notices and nothing below. We also hold occasional notices mentioning about twenty further countries; those are not coverage and we don't count them. The coverage page lists every source, its licence, its jurisdiction and when it last synced.

That's a use we build for, not one we tolerate. Every record carries its source URL, notice identifier and retrieval date, so a finding traces to a primary document. Values stay in the currency they were published in and are never converted or summed across currencies. Contract amendments are marked and kept out of totals, which matters more than it sounds: in one review, amendments accounted for over half of what a naive league table would have called new contract value. Known gaps are published as data rather than buried. For non-commercial newsroom and academic use, talk to us rather than assuming the commercial price.

You check it. Every figure decomposes into records, and every record links to the publisher's own page — that link, not our copy, is authoritative. Beyond that, the accuracy rules are deliberate and documented: pre-award estimates are flagged so a framework ceiling is never reported as money won; a value wildly out of scale with its buyer's history is served as published but excluded from totals; a supplier is never credited on a guess. Where we derive something rather than read it, the record says so.

Three kinds of gap, all of them named on the coverage page. Structural: invitation-only quick quotes and framework further-competitions publish no advertisement anywhere, only their awards — any platform implying otherwise is selling you the gap. Technical: a handful of portal families sit behind CAPTCHA, WAF or JavaScript walls we won't break; their regulated notices still reach us through the national feeds, their below-threshold quotes don't. Legal: some countries publish under NonCommercial licences, which rule out a paid service — Rwanda's and Indonesia's among them — so we don't ingest them, and we'd rather ask permission than assume it. Two sources we do carry, Colombia's SECOP II and Italy's CIG register, are share-alike: their records say so, and anyone may reuse them on the same terms.