legal_mcp

# Fedlex MCP Server — Swiss Federal Law for Claude An MCP server that gives Claude access to Swiss federal law via the [Fedlex](https://www.fedlex.admin.ch/) platform. ## What it does The server exposes 5 tools: | Tool | What it does | Data source | |---|---|---| | `search_laws` | Find laws by keyword in their title | SPARQL → metadata | | `get_law_metadata` | Get detailed info about a specific law | SPARQL → metadata | | `get_law_text` | Fetch the actual text (or a specific article) | SPARQL → URL → HTTP fetch HTML | | `list_amendments` | Show version history of a law | SPARQL → metadata | | `browse_taxonomy` | Navigate the SR classification tree | SPARQL → vocabulary | ### How the Fedlex API works (the two layers) ``` ┌─────────────────────────────────────────────────────────────┐ │ Layer 1: SPARQL endpoint │ │ https://fedlex.data.admin.ch/sparqlendpoint │ │ │ │ Contains: METADATA │ │ - Which laws exist (ConsolidationAbstract) │ │ - Titles, SR numbers, dates, taxonomy │ │ - Links between acts (amendments, basic acts) │ │ - Version history (Consolidation entries) │ │ - FRBR chain → URLs to actual documents │ │ │ │ Does NOT contain: the actual law text │ └──────────────────────┬──────────────────────────────────────┘ │ follow URL from metadata ▼ ┌─────────────────────────────────────────────────────────────┐ │ Layer 2: Document files │ │ https://www.fedlex.admin.ch/eli/cc/... │ │ │ │ Contains: ACTUAL TEXT │ │ - HTML, PDF, XML (Akoma Ntoso) versions │ │ - The real law content │ └─────────────────────────────────────────────────────────────┘ ``` The `search_laws` and `get_law_metadata` tools only use Layer 1. The `get_law_text` tool uses both: first SPARQL to find the URL, then HTTP to fetch the text. ## Setup ### 1. Install dependencies ```bash cd fedlex-mcp pip install -r requirements.txt ``` ### 2. Test it works ```bash python -c " from fedlex_client import search_laws results = search_laws('Raumplanung') for r in results: print(f\"{r['title']} — {r['url']}\") " ``` ### 3. Add to Claude Desktop Edit your `claude_desktop_config.json`: **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json` **Windows**: `%APPDATA%\Claude\claude_desktop_config.json` ```json { "mcpServers": { "fedlex": { "command": "python3", "args": ["/ABSOLUTE/PATH/TO/fedlex-mcp/server.py"] } } } ``` Replace `/ABSOLUTE/PATH/TO/` with the actual path on your machine. Restart Claude Desktop. You should see the Fedlex tools in the connector menu. ### 4. (Optional) Run as remote server for Cowork If you want to share this with your team or use it via Cowork as a remote connector: ```bash python server.py --transport sse ``` Then add the URL (e.g. `http://localhost:8000/sse`) in Claude Desktop under Settings → Connectors → Add custom connector. **Note**: For remote/Cowork, you'll want to deploy this behind HTTPS (e.g. on Cloud Run, Fly.io, or a simple VPS with nginx + Let's Encrypt). ## Example conversations Once connected, you can ask Claude things like: - *"Welche Gesetze gibt es zur Raumplanung?"* - *"Was sagt Art. 22 des Raumplanungsgesetzes?"* - *"Wann wurde das Umweltschutzgesetz zuletzt geändert?"* - *"Zeig mir die SR-Klassifikation für Baurecht"* - *"What does the Swiss federal constitution say about property rights?"* ## Architecture ``` Claude ←→ MCP Protocol ←→ server.py ←→ fedlex_client.py │ ├→ SPARQL queries → fedlex.data.admin.ch └→ HTTP fetch → www.fedlex.admin.ch ``` The MCP server (`server.py`) defines the tools. The client (`fedlex_client.py`) handles the actual SPARQL queries and HTML fetching. This separation makes it easy to test the client independently or reuse it in other contexts. ## Data model (JOLux ontology) The SPARQL data follows the JOLux ontology with FRBR abstraction levels: ``` ConsolidationAbstract (= a law in the SR, e.g. "RPG") │ ├── Expression (title in DE/FR/IT/EN) │ └── Consolidation (= a specific version at a point in time) │ ├── Expression (text in DE/FR/IT/EN) │ │ │ └── Manifestation (HTML / PDF / XML format) │ │ │ └── URL (the actual file) │ └── dateApplicability (when this version applies) ``` Full documentation: https://swiss.github.io/fedlex-jolux/ ## License MIT