The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Parlament MCP listing page.
Part of the Swiss Public Data MCP Portfolio – connecting AI models to Swiss public data sources.
Note: This server covers the federal level (Curia Vista). This repo additionally hosts a self-contained subproject under
openparldata-mcp/— the subnational counterpart for the 26 cantons and ~70 municipal parliaments (OpenParlData.ch). The two are independent servers.
An MCP server that connects AI models to the Swiss Federal Parliament via the
Curia Vista OData API (ws.parlament.ch).
Access motions, interpellations, votes, members, sessions, and the verbatim
debate transcripts of the Amtliches Bulletin – with no API key required
(Phase 1 – No-Auth-First).
Metadata layer:
"Welche Vorstösse zu KI in der Schule sind hängig?" →
parlament_search_business(keyword="KI", keyword2="Schule", status="Eingereicht")
Verbatim transcripts (Amtliches Bulletin):
"What did National Councillor Munz say in the 2024 spring session about the Volksschule? Give me the exact wording with a correct AB citation." →
parlament_search_transcripts(speaker_name="Munz", session_id=5202, keyword="Volksschule")→ thenparlament_get_transcript(transcript_id=…)for the full wording.Returns short, citable excerpts (
AB 2024 N, 2024-03-13, Munz Martina) with a stable source URL; the verbatim text is fetched on demand, never in bulk.
Perfect for the KI-Fachgruppe Stadtverwaltung Zürich: find pending motions on AI in education, or quote what was actually said in the chamber – instantly.
| Tool | Description |
|---|---|
parlament_search_business | Search Vorstösse by keyword, type, status, council, date |
parlament_get_business | Full details of a single business (texts, FC response) |
parlament_search_members | Find councillors by canton (e.g. ZH), party, council |
parlament_get_votes | Parliamentary votes with Ja/Nein meaning |
parlament_get_sessions | List recent sessions with IDs for follow-up queries |
parlament_search_transcripts | Search debate transcripts → citable excerpts with AB citation + source URL (speaker / session / business / date filters) |
parlament_get_transcript | Fetch the verbatim full text of a single speech by ID (capped, paginated, is_excerpt flagged) |
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
By default the server binds to 127.0.0.1 (localhost only). Set
MCP_HOST=0.0.0.0 only inside a container/cloud context (Docker, Railway,
Render, Kubernetes). Never bind to 0.0.0.0 on a local dev machine – it exposes
the server to your local network (NeighborJack risk); the server logs a warning
if you do so outside a detected container.
Transport is selected via MCP_TRANSPORT (stdio default, or sse /
streamable-http); --http is kept as an alias for streamable-http.
The HTTP transport is open by default (public read-only data). To require a
bearer token, serve via the CORS/auth app factory and set MCP_BEARER_TOKENS:
Each request then needs Authorization: Bearer <token>; identity comes from the
validated token, not a session header (see docs/security.md).
Kubernetes manifests (hardened securityContext, resource limits, egress
NetworkPolicy, Mcp-Session-Id sticky routing) live in deploy/k8s/;
an HAProxy stick-table example is in deploy/haproxy.cfg.
| Partner Server | Combination |
|---|---|
openparldata-mcp | Federal ↔ subnational — same question across cantons & municipalities |
fedlex-mcp | Law text ↔ parliamentary debate that created it |
zurich-opendata-mcp | City policy ↔ cantonal/federal motions |
swiss-statistics-mcp | Data backing ↔ motions citing statistics |
Power query example:
Official proceedings of Swiss authorities are excluded from copyright under Art. 5 para. 1 lit. a URG (Swiss Copyright Act). The verbatim wording of parliamentary debates in the Amtliches Bulletin may therefore be reproduced and quoted freely — which is exactly what the transcript tools return: the wording itself, never a summary standing in for it.
| Source | License | Attribution |
|---|---|---|
| Curia Vista (ws.parlament.ch) | CC BY 4.0 | © Schweizer Parlament, CC BY 4.0 |
Every tool returns a typed structured response (FastMCP exposes the output
schema) carrying source, license, provenance, match_type and count
alongside typed results. Data is passed through unmodified.
This server is in Phase 1 — Read-only Wrapper (all tools readOnlyHint: true,
no writes). The full phase model and transition criteria are in
docs/roadmap.md.
This server speaks two protocol eras over the same endpoint. The client's first request on a connection decides which one applies; a later claim from the other era is refused.
| Era | Revision | Who reaches it |
|---|---|---|
initialize handshake | 2024-11-05 … 2025-11-25 | What today's clients speak. The server answers with the revision asked for, or with the 2025-11-25 ceiling when the request asks for something newer. |
| Per-request envelope | 2026-07-28 | A request carrying the 2026-07-28 _meta envelope opens a modern connection. |
Both revisions are pinned in
tests/test_protocol_version.py and asserted
against the installed SDK, so a Dependabot bump of mcp cannot move either one
silently. This server builds no ASGI app to send an initialize through, so
the gate asserts the SDK constants rather than a measured response — the
weaker form, named rather than left unsaid.
Note that the SDK's LATEST_PROTOCOL_VERSION is an alias for the modern
era, not for the handshake era — pinning against it alone would leave the era
that current clients actually negotiate free to drift.
Update policy. When the gate fails, do not edit the constant blindly: read
the spec changelog between the two revisions, verify the server still behaves,
then move the constant, this section, README.de.md and
CHANGELOG.md together.
Phase 1 uses Tools only — rationale and the Phase-2 Resources plan are in
docs/adr/ADR-003-mcp-primitives.md.
All tools declare explicit annotations consistent with their behaviour:
| Tool | readOnly | destructive | idempotent | openWorld |
|---|---|---|---|---|
parlament_search_business | ✅ | — | ✅ | ✅ |
parlament_get_business | ✅ | — | ✅ | ✅ |
parlament_search_members | ✅ | — | ✅ | ✅ |
parlament_get_votes | ✅ | — | ✅ | ✅ |
parlament_get_sessions | ✅ | — | ✅ | ✅ |
parlament_search_transcripts | ✅ | — | ✅ | ✅ |
parlament_get_transcript | ✅ | — | ✅ | ✅ |
Structured JSON logs go to stderr (stdout stays reserved for the stdio
protocol). OpenTelemetry tracing wraps each tool call and auto-instruments
outgoing HTTP; set OTEL_EXPORTER_OTLP_ENDPOINT (with the otel-export extra)
to ship spans. See docs/security.md for the full security
posture (Lethal-Trifecta assessment, egress allow-list, gateway hardening).
| Aspect | Details |
|---|---|
| Access | Read-only (readOnlyHint: true) — the server cannot modify or delete any data |
| Personal data | Parliamentary businesses are public record by law (BGÖ). No private data is accessed or stored. |
| Rate limits | Built-in per-query caps: max. 100 results (businesses/members), 50 (votes), 30 (transcript search), 10 (sessions). Transcript full text is capped per call and paginated — you never pull a whole session by accident. |
| Timeout | 20 seconds per metadata call; 45 seconds for transcript reads (verbatim search is heavier) |
| Authentication | No API keys required — Curia Vista is publicly accessible |
| Data source | Official Swiss federal government data (Schweizerische Parlamentsdienste) |
| Terms of Service | Subject to ToS of ws.parlament.ch — Schweizerische Parlamentsdienste |
General
substringof() filter is case-sensitive for some fields.null in the API for very recent sessions – use session ID.Transcripts (Amtliches Bulletin) — verified live 2026-07-19:
Transcript entity
reaches back to Dec 1999. Debates from 1891–1999 exist only as scanned
archive documents (Bundesarchiv / Amtsdruckschriften) and are not
connected here (no OCR in scope). A query whose whole date window predates
coverage returns an explanatory error, not an empty result.AB <year> N <page> form cannot be built. We emit a stable, verifiable
substitute — AB <year> <N|S>, <date>, <speaker> — plus the authoritative
source_url (SubjectId) and the transcript_id. This is a documented,
honest trade-off, not an omission.Language is the edition, not the spoken
language. The tools filter Language eq 'DE' purely to deduplicate the three
byte-identical editions (DE/FR/IT) down to one copy. Every speech is returned in
its original wording; a French or Italian speech is not hidden — its real
language is reported in the language field (de/fr/it). The response
states this via language_note.snippet, ~320 chars) with is_excerpt, total_length_chars and a hint to
fetch the full text. parlament_get_transcript caps output at max_chars; when
it truncates it sets is_excerpt=True and returns a next_offset to continue.keyword combined with a speaker_name is the slowest
path (~40 s). Add a session_id, business_number or date window to keep reads
around 1–2 s.HTTP is mocked with respx; network-dependent tests are marked
@pytest.mark.live and excluded from CI via -m "not live". Tool definitions are
pinned in tool-hashes.json (python -m parlament_mcp.tool_hashes --check).
See CONTRIBUTING.md.
See SECURITY.md for the security policy and posture (vulnerability reporting, Lethal-Trifecta assessment, accepted risks).
MIT © Hayal Oezkan — see LICENSE
Hayal Oezkan · github.com/malkreide
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):