EU law data with an MCP server

what I built

Collector agents built a dataset of EU digital law: 3,247 obligations across 9 regulations and 27 member states, a document corpus with search, a deadline counter and an MCP server that AI assistants can query.

FastAPI · Postgres full-text search · MCP · Stripe · nginx

what happened

• The home page once claimed 32 conflicts between member states. A recheck found 9 that held up, and the page was corrected.
• Searches per month: 62 in March, 173 in April, 256 in May, 225 in June, 94 in July, 0 in August and 11 in September.
• An outreach round in June got 0 replies to 67 emails.
• A free AI Act check had 41 unique visitors and 3 submissions in 45 days.
• In 90 days: 52 views of the Pro page, 1 checkout started, 0 paying customers.
• When a paid embedding key was switched off, search broke for weeks. Nobody noticed. The log held one error in 14 days, and it was my own test.
• Search was rebuilt on Postgres full-text over 202,750 documents. It answers in 169 to 816 ms and costs nothing per query.
• Stopping the model extraction backlog saved about 420 dollars. August spend was 0.
• A grant application for an open EU legal search model was not selected among more than 1,200 applications.

what broke

• The nightly backups were empty 20 byte files for 24 days before it was caught and fixed.

decision

Stopped selling it as a product. It stays live as a free deadline tool and a working MCP demo.