🏛️ lobbywatch-mcp

An MCP server that connects AI models to Lobbywatch.ch, the largest lobby database of the Swiss Federal Parliament — conflicts of interest, lobby groups, access badges, and transparency scores.

🇩🇪 Deutsche Version

Part of the [Swiss Public Data MCP Portfolio](https://github.com/malkreide) — connecting AI models to Swiss public data sources.

🎯 Anchor Demo Query

"Welche Mitglieder der WBK-N haben Interessenbindungen zu Bildungsverlagen oder privaten Bildungsträgern, und wie ist ihre Transparenz-Bewertung?"

Which members of the National Council's Education Commission have declared conflicts of interest with educational publishers or private education providers, and how does their compensation transparency score compare?

→ More use cases by audience →

Overview

Lobbywatch.ch maintains the largest public database on Swiss federal parliamentarians and their connections to lobby organisations: 245 parliamentarians, ~7'800 interessenbindungen (declared mandates), 139 lobby groups, 368 access-badge holders, updated weekly, licensed CC BY-SA 4.0.

lobbywatch-mcp exposes this data to Large Language Models via the Model Context Protocol. It is designed to be used alongside parlament-mcp (the official Swiss Parliament's Curia Vista data): the pair makes it possible to ask what a parliamentarian did officially and who they are connected to — in a single conversation.

Features

• Dump-first, API-fallback architecture. The weekly JSON dump is the primary source (stable, verified in production); the live dataIF REST API is used only where it returns reliable data (lobby groups, search).
• Seven Phase 1 tools — parliamentarian lookup, conflict-of-interest listing, branche search, lobby group fetch, rankings, transparency quota, cache control.
• CC BY-SA 4.0 attribution baked into every response via Pydantic envelopes.
• Dual transport — stdio for Claude Desktop, streamable-http / sse for cloud deployments.
• Fuzzy name matching via rapidfuzz for natural LLM input like "Jositsch" or "Wehrli".
• No authentication required (Phase 1 — No-Auth-First).

Architecture

                    ┌─────────────────────────────┐
   LLM client       │     LobbywatchClient        │
  (Claude Desktop,  │                             │
   Inspector, …)    │   ┌───────────────────┐     │
        │           │   │  Dump cache       │     │      cms.lobbywatch.ch
        │  MCP      │   │  (24 h TTL,       │     │      ┌──────────────────┐
        ▼  stdio /  │   │   ~80 MB resident)├─────┼─────►│ weekly JSON      │
   ┌─────────┐ HTTP │   └───────────────────┘     │      │ export (~17 MB)  │
   │ FastMCP │◄────►│                             │      └──────────────────┘
   │ server  │      │   ┌───────────────────┐     │      ┌──────────────────┐
   └─────────┘      │   │  dataIF REST      ├─────┼─────►│ /interface/v1/   │
                    │   │  (live fallback)  │     │      │   json/…         │
                    │   └───────────────────┘     │      └──────────────────┘
                    └─────────────────────────────┘

Outbound HTTP runs through a single httpx.AsyncClient with follow_redirects=False, an SSRF guard that blocks RFC1918 / link-local / metadata IPs, and an httpx event hook that re-resolves on every request. The dump path is the primary source of truth for parliamentarian queries; dataIF is only used for lobby group lookups and the search endpoint.

Prerequisites

• Python 3.11 or newer
• Internet access to download the weekly Lobbywatch JSON export (~17 MB zipped)

Installation

From PyPI (after first release):

pip install lobbywatch-mcp

From source:

git clone https://github.com/malkreide/lobbywatch-mcp.git
cd lobbywatch-mcp
pip install -e ".[dev]"

Usage

Standalone

lobbywatch-mcp

This starts the server in stdio mode. For HTTP:

LOBBYWATCH_MCP_TRANSPORT=http LOBBYWATCH_MCP_PORT=8000 lobbywatch-mcp

Container

A hardened multi-stage Dockerfile ships with the repo (non-root, read-only-rootfs compatible). See docs/deployment.md and deploy/docker-compose.example.yml for resource limits, sticky-LB guidance and egress hardening.

docker build -t lobbywatch-mcp:0.2.0 .
docker run --rm -p 127.0.0.1:8000:8000 lobbywatch-mcp:0.2.0

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "lobbywatch": {
      "command": "uvx",
      "args": ["lobbywatch-mcp"]
    }
  }
}

A full example is provided in claude_desktop_config.json.

Example Queries

Once connected, try prompts such as:

• "Give me the top 10 parliamentarians by number of interessenbindungen in the SP party."
• "Which WBK-N members have mandates in the publishing or education industry?"
• "Look up the lobby group 'economiesuisse' and list its connected parliamentarians."
• "What is the compensation-transparency score distribution for the finance commission (FK-N)?"

Tools

All tool names use the lobbywatch_ namespace prefix (since 0.2.0) to avoid collisions with sibling portfolio servers.

Configuration

All behaviour is controlled via environment variables:

Project Structure

lobbywatch-mcp/
├── src/lobbywatch_mcp/
│   ├── __init__.py
│   ├── __main__.py        # CLI + transport selection
│   ├── config.py          # URLs, cache paths, attribution
│   ├── client.py          # Dump download + dataIF client
│   ├── models.py          # Pydantic v2 response envelopes
│   └── server.py          # FastMCP tool registrations
├── tests/
│   ├── conftest.py        # Fixture parliamentarians
│   ├── test_client.py     # Respx-mocked unit tests
│   ├── test_server.py     # Tool integration tests
│   └── test_live.py       # @pytest.mark.live — excluded from CI
├── .github/workflows/
│   ├── ci.yml             # Test matrix + ruff
│   └── publish.yml        # PyPI OIDC Trusted Publisher
├── claude_desktop_config.json
├── pyproject.toml
└── ...

Data License & Attribution

The code is released under the MIT License.

The data served through this MCP is © Lobbywatch.ch and licensed under CC BY-SA 4.0. Every response envelope includes the attribution string. Downstream users must:

1. Credit Lobbywatch.ch as the data source.
2. Share derivative datasets under the same CC BY-SA 4.0 terms.
3. Understand that Lobbywatch is a community-researched database — not an official register. It is authoritative for transparency research but should not be confused with the Federal Parliament's own declarations.

Known Limitations

• The upstream /table/parlamentarier/... dataIF REST endpoint currently returns empty result sets. The server works around this by using the weekly JSON dump instead.
• zutrittsberechtigungen (access badges) are not populated in the "essential" dump variant used here. A future release will add a dedicated tool using the non-essential dump.

Contributing

See CONTRIBUTING.md.

Security

See SECURITY.md for the security posture, accepted-risk decisions, and how to report a vulnerability.

Changelog

See CHANGELOG.md.

License

MIT License — see LICENSE. Data CC BY-SA 4.0 — see LICENSE § Data notice.

Author

malkreide · GitHub

Part of the [Swiss Public Data MCP Portfolio](https://github.com/malkreide).

<!-- mcp-name: io.github.malkreide/lobbywatch-mcp -->

<!-- BEGIN GENERATED: install -->

Installation

Run via uv's uvx — no clone or manual install needed. Add to your MCP client config (mcpServers for Claude Desktop, Cursor and Windsurf; use a top-level servers key for VS Code in .vscode/mcp.json):

{
  "mcpServers": {
    "lobbywatch-mcp": {
      "command": "uvx",
      "args": [
        "lobbywatch-mcp"
      ]
    }
  }
}

<!-- END GENERATED: install -->