· via Hacker News – Front Page (native)
Pandex gives AI agents search access to 2.8M official European legal documents over MCP
Pandex offers a single REST API and MCP server over 2.8 million official EU and national legal documents, with multilingual search and a public register of coverage gaps.
A new service called Pandex, which surfaced on the front page of Hacker News, is offering a single REST API and MCP server that makes 2.8 million official European legal documents searchable by AI agents and ordinary software. The corpus pulls legislation and highest-court decisions directly from the institutions that publish them, covering the European Union and seven European countries.
What the service provides
According to Pandex, one API and one MCP server cover every jurisdiction the company holds, returning results as Markdown or JSON. Search can be keyword-based, semantic or a hybrid of the two, and it works across languages — the company's own example is asking a question in English and retrieving a French Cour de cassation ruling on whether a bike courier for a delivery app counts as an employee, along with the passage that answers it.
The headline numbers: 2.4 million court decisions, 421,000 acts and codes, 112 million searchable passages and 37 million citations linked back to their source text. The MCP server integrates with agent tooling including Claude, Cursor, GitHub Copilot, Windsurf, Zed and Gemini. Pandex pitches this at three audiences: legal AI products that need answers grounded in real law with citations, compliance teams tracking changes to the law across jurisdictions with past versions and effective dates, and legal research products that would otherwise maintain a separate scraper for every court.
Coverage, and the gaps it publishes
Coverage currently spans EU institutions plus France, Germany, the Netherlands, the United Kingdom, Austria, Switzerland, Luxembourg and the Council of Europe's human rights court, with Spain, Italy, Ireland, Belgium, Poland, Czechia, Estonia and Finland listed as next.
What stands out is the transparency. Pandex publishes an institution register, dated 17 November 2026, showing 33 of 41 institutions held complete, and it documents exactly what is missing and why. The French Cour de cassation is partial because decisions the court publishes only through Judilibre since 2020 require a registered account, and the official dumps run about two months behind. Germany's Federal Constitutional Court is partial because chamber decisions from 1998 to 2009 exist only on the court's own site, which refuses clients that are not browsers. In the Netherlands, roughly 27,700 older decisions of the Council of State have no text at the official Rechtspraak portal, and the council's own site sits behind a Cloudflare challenge. The UK's House of Lords Appellate Committee is not held at all: about 900 judgments from 1996 to 2009 sit on archived parliament.uk pages the Pandex server cannot reach, and earlier ones were never published at an official site. Luxembourg's Cour de cassation is partial because older judgments sit behind a proof-of-work challenge, while the European Court of Human Rights' Treaty Office blocks the company's server with Cloudflare. Some collections, such as Germany's Federal Administrative Court and Switzerland's federal courts, are simply still loading.
Pricing
A free tier allows 1,000 requests a day across the API and MCP server, with access to everything Pandex holds in every jurisdiction. Paid usage is priced at €1 per 1,000 requests, aimed at products with their own users. There is also a corpus-delivery option that ships every text, version and citation link as files, plus each day's changes, so customers can store, index and build on the data in their own systems. Pandex is offering ten founding-customer places at €300 a month, fixed for two years, which include up to 1 million requests a month without per-request counting and the right to name the next country or source the company loads.
Why it matters
Two things make this more than another legal search engine. First, the MCP delivery model: agent ecosystems have a grounding problem, and legal reasoning is where hallucinated citations do the most damage. A structured, citable corpus reachable directly from tools like Claude or Cursor gives agents a verifiable path from an answer back to the source text. Second, the fragmentation it papers over: European legal publishing is a patchwork of formats, delayed dumps, archived sites and anti-bot walls, which until now has pushed every legal tech product into building and maintaining its own scrapers. A single schema across jurisdictions removes that barrier — and the decision to publish what is not held, rather than hide it, lets builders know exactly when their agent might be missing the one judgment that matters.
- #mcp
- #ai-agents
- #legal-tech
- #api
- #european-union