LawDB

Connect via MCP

LawDB exposes German and EU legislation plus case law from German and EU courts, the International Court of Justice, and the Permanent Court of Arbitration as an MCP server. A language model can search it, retrieve full texts including available paragraph numbers, and read individual provisions.

Endpoints

Full tool set (Claude, Claude Code, generic clients)
https://lawdb.eu/mcp
ChatGPT connector (search / fetch)
https://lawdb.eu/mcp/chatgpt

Claude: sign in with OAuth

Add the endpoint in Claude as a connector. Claude discovers the sign-in flow on its own, shows a consent screen and stores the token. A LawDB account is all you need.

ChatGPT and scripts: sign in with a token

Sign in to LawDB to create your API key (personal access token). Go to Account and API keys Create a token there with the "mcp:use" ability and supply it as a bearer token. 

Authorization: Bearer <your token>

What the server does

  • search_legal_documents — Search legislation and case law, filtered by jurisdiction, document type, judgment vs Opinion, court and date range. Full-text and vector search run together.
  • get_document — Full text of a decision or act in the language you pick, with paragraph numbers preserved and the AI summary where one exists.
  • search_provisions — Search inside a single act, named by short name, ELI, CELEX or the id of a search result. Returns every matching provision with its link, page by page — and for acts held as one continuous text, the passages in the wording.
  • get_provision — A single provision, addressed by its official marker.
  • list_jurisdictions — Valid filter values: jurisdictions, languages, courts.

Notes on citing

  • Paragraph numbers appear as [Rn. 42] in the text and are the citation unit. Where a source carries none, the response says so explicitly.
  • AI summaries exist in German or English only, regardless of the language of the text. They are machine-generated and validated against verbatim quotations from the source.
  • EU documents exist in up to 24 language versions; every response lists the ones held.