convez.ai
MCP ServerIntegração & APIsIntermediário
HeinzTempl

RIS-MCP: Pesquisa de Direito Austríaco

HeinzTempl/ris-mcp

Pesquisa jurídica austríaca local para Claude e LLMs

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP local que integra a API aberta de legislação e jurisprudência austríaca (RIS-OGD) ao Claude e outros clientes MCP. Fornece acesso consolidado a leis federais, estaduais e decisões de tribunais superiores austríacos em Markdown limpo, sem necessidade de chave de API.

Instalação

$npx convez-ai HeinzTempl/ris-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
2
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Pesquisar legislação federal e estadual austríaca consolidada
  • Buscar jurisprudência de tribunais superiores (OGH, VfGH, VwGH)
  • Correlacionar normas legais com decisões jurisprudenciais relevantes
  • Análise jurídica com garantia de precisão (dados brutos da API RIS)

Veredito

Ferramental especializado e bem executado para pesquisa jurídica austríaca com excelente atenção a precisão e privacidade. Ideal para advogados e aplicações legais que funcionam com modelos locais ou Claude, mas com escopo geográfico restrito ao direito austríaco.

README

A tradução para o português deste README ainda está na fila.

ris-mcp

Python MCP Austrian law License: MIT

A local MCP server for Austria's RIS Open-Government API (data.bka.gv.at/ris/api). It gives Claude, Msty, LM Studio and other MCP clients tools for federal law (consolidated), state law (consolidated) and case law (OGH, VfGH, VwGH, BVwG, …). Every response comes back as clean Markdown — ready for a language model and readable by a human.

The server is built for local models: it bundles the typical research chain into single calls so even small models don't have to orchestrate many tool calls, and it is tuned to make hallucinating legal content hard — every case number, date, document ID and headnote is taken verbatim from the API.

In practice, you ask in natural language and the assistant picks the right tool:

„Was sagt § 1295 ABGB und welche OGH-Judikatur gibt es dazu?" „OGH-Entscheidungen zu Hund Schmerzengeld." „Volltext von § 19 Wiener Bauordnung."

No API key required — the RIS-OGD interface is open. The server caches locally and rate-limits its requests to be a good citizen.

Scope

The focus is Austrian civil law and the OGH — that's where the workflow is most polished. The same tools also reach the other courts (VfGH, VwGH, BVwG, LVwG, DSB, plus OLG/LG via the Justiz application) and the consolidated state law of all nine federal states.

Tools

ToolPurpose
ris_recherche_normOne-call research: full text of a provision plus the most relevant OGH headnotes, in one document. The easiest entry point for "case law on § X".
bundesrecht_searchSearch the consolidated federal collection (BrKons).
bundesrecht_get_normFull text of a single section / article / annex.
bundesrecht_get_gesamte_vorschriftA whole statute (e.g. the complete ABGB) as HTML→Markdown.
landesrecht_searchSearch consolidated state law (LrKons); use the bundesland filter.
landesrecht_get_normFull text of a single state-law provision.
ogh_searchOGH search, optimised for the civil-law workflow. Searches headnotes and decision texts automatically and falls back to Austrian spelling (e.g. Schmerzengeld).
judikatur_searchGeneric search for every other court (gericht enum). Same auto two-mode behaviour as ogh_search.
judikatur_get_entscheidungFull text of a headnote or decision document.
ris_local_searchFTS5 search over locally cached documents (no RIS call).
ris_healthServer and cache status.

Works with

Any MCP-compatible client:

ClientModelPrivacy
Claude Desktop (Anthropic)CloudQuery text goes to Anthropic
ChatGPT Desktop (OpenAI)CloudQuery text goes to OpenAI
Msty StudioCloud or localFully local possible
LM StudioLocalStays on your machine
Mistral VibeCloud or localDepends on configuration
Any other MCP-compatible client——

RIS-OGD is public open data — no client documents touch this server. The privacy column refers only to what the LLM (cloud or local) sees of the query text you type.

Note on local models: the model must support reliable tool calling. Qwen3 6B+ and Gemma 4 work well; smaller models (under 4B) may struggle.

Installation

git clone <repo-url> ris-mcp
cd ris-mcp
uv sync

Requires uv and Python ≥ 3.12.

MCP client configuration

In your client's MCP config (e.g. Msty: Settings → MCP, or an mcp.json):

macOS / Linux:

{
  "mcpServers": {
    "ris": {
      "command": "/opt/homebrew/bin/uv",
      "args": ["run", "--directory", "/Users/you/projects/ris-mcp", "ris-mcp"]
    }
  }
}

Windows (forward slashes work; backslashes must be escaped as \\):

{
  "mcpServers": {
    "ris": {
      "command": "C:/Users/you/.local/bin/uv.exe",
      "args": ["run", "--directory", "C:/Users/you/projects/ris-mcp", "ris-mcp"]
    }
  }
}

Adjust paths and restart the client — all tools then appear under the server ris.

Use the absolute path to uv. Apps launched from the Dock (macOS) or the Start menu (Windows) get a minimal PATH, so a bare "uv" may fail to start the server intermittently. Find the absolute path with:

  • macOS / Linux: which uv
  • Windows (PowerShell): (Get-Command uv).Source
  • Windows (cmd): where uv

Examples

Provision + case law in one call:

ris_recherche_norm(gesetz="ABGB", paragraph=1295)

returns the full text of § 1295 ABGB together with the most relevant OGH headnotes (each with its headnote text and doc_id). For the full reasoning of a decision, follow up with judikatur_get_entscheidung(doc_id="…").

Topic / fact-pattern search:

ogh_search(suchworte="Hund Schmerzengeld")

returns recent OGH decisions on the topic, newest first. Keep search terms sparse — they are AND-combined, and the concrete facts ("Hund", "gebissen") live in the decision text, not in the abstract headnote.

Research system prompt

A matching system prompt — recherche-systemprompt.md — ships with the server, for use with a local model. It enforces a strict anti-hallucination discipline: no legal content from the model's own knowledge, only verbatim from RIS.

Cache

A single SQLite file, located via platformdirs:

  • macOS: ~/Library/Caches/ris-mcp/cache.db
  • Linux: ~/.cache/ris-mcp/cache.db
  • Windows: %LOCALAPPDATA%\ris-mcp\Cache\cache.db

Three areas: search_cache (search results, 24 h TTL), document_cache (full texts, 30-day TTL, historical versions kept permanently) and meta (HTTP ETag / Last-Modified per URL for conditional requests). An FTS5 index over the cached texts backs ris_local_search. The cache lives outside the repo; delete the file to reset.

Settings can be overridden via RIS_MCP_* environment variables (see src/ris_mcp/config.py).

Note

This is a private research-support tool, not a substitute for legal review. The authoritative text is always the official RIS original. Assessment and subsumption remain the task of the lawyer.

License

MIT.

Ver o original no GitHub

Bom para

Relacionados