Swissmedic MCP server

@pipeworx/swissmedic

Swissmedic's authorised human medicines register (Switzerland) — the "Zugelassene Arzneimittel HAM" list: every authorised product and each of its authorised dosage strengths, with the marketing-authorisation holder, medicine category, dispensing category (prescription tier) and authorisation dates. Keyless.

Part of Pipeworx — an MCP gateway connecting AI agents to 1683+ live data sources.

Tools

• swissmedicsearch(query?, holder?, category?, dispensingcategory?, limit?, offset?) — free-text/filtered search with real offset+limit pagination and a total_matches count. Answers "is X authorised in Switzerland", "who holds the Swiss authorisation for X", "what dispensing category is X in".
• swissmedicproduct(name) — every authorised dosage strength for one product by name (exact match preferred, falls back to substring with matchtype marked). The export carries no indication text and no per-product URL, so this returns the authoritative Swissmedic download instead.
• swissmedicrecentchanges(since) — products first authorised, or with a dosage strength newly authorised, since a date — a real signal from the source's own date columns.
• swissmedicdatasetinfo() — the authoritative bulk-download URL, source page, row count, and the full lists of distinct medicine categories and dispensing categories used by the filters.

Auth

Keyless.

Data sources

• — the authorised human medicines XLSX, linked from swissmedic.ch's "Listen und Verzeichnisse" page. Verified live 2026-09-25: HTTP 200, 1,005,371 bytes, 11,086 data rows. Swissmedic states the lists are "updated monthly" — there is no finer-grained cadence published.

This pack fetches EMA-sized data, not ema-article57-sized data — and that difference decides the architecture

The file is ~1 MB compressed, ~4.8 MB of sheet XML. That is the same size class the sibling ema-medicines pack already parses live per request without issue, and roughly 9x smaller than the file that made the ema-article57 pack build a local R2-backed copy instead (164,314 rows / 63 MB of sheet XML there, versus 11,086 rows / 4.8 MB here). So swissmedic fetches live, same pattern as ema-medicines: one parsed copy held per isolate for 12 hours (DATASETTTLMS) plus a 12-hour Cloudflare edge cache on top — Swissmedic's own monthly update cadence makes anything shorter needless re-downloading of an unchanged file. If a refresh fails while an earlier copy is in hand, the earlier copy is served rather than an error.

Licence — this is a live proxy, not a mirror, so no separate reuse grant is needed

Swissmedic's general Legal Notice is more restrictive-sounding than EMA's ("downloading or copying content... requires written consent from the copyright holders" — a page written for the site broadly, not this specific public list). That question does not need resolving here: per the proxy-vs-mirror rule (mirror-needs-grant-proxy-does-not, CLAUDE.md), a pack that calls a public, keyless upstream live on each cache-miss is a client, not a publisher, and needs no explicit grant — only bulk-copying/mirroring the dataset would. This pack does the former (fetch-and-cache, same as ema-medicines), never the latter.

Schema — row 7 is the header, bilingual DE/FR source

Shortened. The full README is on GitHub.