Reliable legal context for AI

Catch AI hallucinations before they become a risk

2 of 8 citations not in the corpus

Klaracase checks the case citations in AI drafts against its own corpus in real time, as an MCP server or REST API.

Paste any legal text. Klaracase checks the case citations in it against the corpus, live. § references are detected and marked not yet verifiable. What holds is confirmed. What does not is flagged.

01 · 3 places it runs · one legal corpus

One check. Everywhere decisions get made.

Klaracase checks cited case-law references against your corpus in real time – in the inbox, before a draft goes out, and directly inside your AI workflows.

Inbound triage

Spot hallucinated references in incoming mail

Some claims and cease-and-desist letters cite decisions that don't exist. Klaracase checks every cited reference against your corpus as mail arrives – so your team never wastes time on a letter with no real basis.

Check inbound mail
02 · Infrastructure

Every question gets its own reference network, not just a list of hits.

Klaracase reads your question by wording and by meaning, both at once, across a corpus of more than 510,000 decisions. The closest passages arrive with their already-resolved reference network, and that becomes the context package for this one question.

More than 1,900 courtsCase law since 1946More than 14M references

  1. 01Question

    Your legal question in plain words

    you ask the way you would in a memo, not in search operators

  2. 02Hits

    The closest passages

    from the entire corpus, not from a pre-sorted selection

  3. 03Citation

    Down to the paragraph number

    every statement backed by court, docket number and paragraph

  4. 04Reference network

    Cited decisions and provisions

    already resolved in the corpus: BGH, XI ZR 12/09 cites § 204 (1) no. 3 BGB

  5. 05Context

    Sourced context, ready to use

    for your file, or over the API for your agent

The corpus in detail
03 · Legal Discovery Graph

Intelligent research, reimagined

State-of-the-art graph and database technologies interlink the German legal corpus at a deep level, surface hidden connections, and provide unique context for demanding expert research.

Reference types: cited, decisive, procedural
Checked against the corpus, not the open web
Why anchoring matters

Answers you can hand to a court, not just to a chatbot.

Legal research fails quietly: a citation that looks right, a paragraph that has been overturned, a source that never existed. The cost surfaces late: in a brief, in a hearing.

Klaracase is built the other way around: every claim enters the system anchored to its source, and everything downstream preserves that anchor.

Measured: without sources, three AI models cited 168 of 374 case numbers we could not confirm. With our data, 1 of 471.
See the benchmark
04 · Features

What Klaracase is built on

Four capabilities, one foundation: every result stays traceable to its source. Each draws on the same structured legal data, German case law and statutes linked into a citation graph, so you get an answer you can check down to the exact source, not a plausible guess.

All features
06 · Connect

Klara MCP

Connect your agents, ChatGPT, or Claude to our MCP server to demonstrably reduce hallucinations in what they write. Your own applications can use the same operations through the REST API.

In our benchmark, with our data: 1 of 471 case numbers not confirmed
Credit-metered, token-authenticated
OpenAPI spec + MCP tool schemas + a field-by-field tool guide
Klara MCP and API in depth
Built and operated in Germany
Server location: GermanyDedicated servers hosted by Hetzner in Germany. Backups stay in Frankfurt (eu-central-1).
GDPR-compliantData processing under Art. 28 GDPR agreements. Every recipient is disclosed in the privacy policy.
Built in GermanyBuilt for German law: court decisions, statutes and citations from German sources.

Start researching on verifiable ground

FAQ

Questions we hear often