The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Openfoodtox listing page.
EFSA's derived toxicological reference values for chemicals in food and feed — acceptable daily intakes, acute reference doses, tolerable intakes and operator exposure levels — with the assessment body and critical endpoint behind each one.
Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.
3,437 substances, 8,620 reference values.
OpenFoodTox 3.0, Zenodo record 19388272
(DOI 10.5281/zenodo.19388272), published 2026-04-30, CC-BY-4.0.
Extracted from EFSA's official .xlsx export — nothing is scraped. The data
ships inside the pack rather than being fetched per call, so answers do not
depend on Zenodo being reachable. Every response carries the DOI, publication
date and licence.
1. Hazard profile. "What is the ADI for glyphosate?" →
openfoodtox_hazard_profile({substance: "glyphosate"}) → every reference value
EFSA has derived, each with its unit, assessment body and critical endpoint.
2. Find the substance. "Which aflatoxins has EFSA assessed?" →
openfoodtox_search({query: "aflatoxin"}) → matching substances and how many
reference values each carries.
One substance can have several legitimate values. Cadmium carries a tolerable weekly intake of 2.5 µg/kg bw from EFSA and 7 from the older JECFA assessment. Both are real. The pack lists them with their assessment bodies rather than reconciling them — picking one would invent a consensus that does not exist.
Absence is not zero. A substance with no derived reference value returns
found:false with an explicit statement that this is not a finding of safety.
An empty list here would read as "no limit applies" to a chemical in food.
Re-run the extractor against a newer Zenodo record and re-run the tests. The tests assert specific published values (glyphosate ADI 0.5 mg/kg bw/day, aspartame 40, cadmium TWI 2.5) precisely so a refresh that breaks the join fails loudly instead of quietly changing numbers.
Column layout worth knowing if you touch the extractor: reference-value columns
are IUCLID dotted paths with a nested element between the metric and the field
(HumanHealthHazardCharacteristics.AcceptableDailyIntake.Adi.lowerValue), and
that element differs per metric. Match on the tail of the path, not the composed
name — matching the composed name parses every row, joins every key, and yields
an empty dataset.
Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):
tools/list at https://gateway.pipeworx.io/openfoodtox/mcp returns the tools in the table
above plus the shared Pipeworx meta-tools — ask_pipeworx,
discover_tools, search_within, remember/recall and the rest of the
gateway-wide set. So the tool count you see is larger than this table: a
single-pack endpoint currently lists roughly 30 shared tools alongside the
pack's own. The connection's initialize response states its exact scope, and
is the authoritative answer for a given day.
This is deliberate, not multiplexing by accident. The meta-tools are what let a
scoped connection answer a question this pack does not cover — via
ask_pipeworx, which routes across the whole catalog — without you adding a
second MCP server. There is currently no way to mount a pack endpoint without
them; if the extra schemas cost you more context than the routing is worth,
connect to the full gateway once rather than to several pack endpoints.
Or connect to the full Pipeworx gateway to get every pack's tools listed directly, instead of just this one's:
Both URLs reach the same gateway and the same 1679+ data sources. The
only difference is which pack's tools are listed directly; ask_pipeworx
reaches all of them from either one.
No account needed for the first calls. Inspect any tool: GET https://gateway.pipeworx.io/v1/tools/openfoodtox_hazard_profile. Find one: POST https://gateway.pipeworx.io/v1/tools/search_packs with {"query":"..."}.
This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:
Or run it directly to confirm it starts:
It speaks MCP over stdin/stdout and answers initialize/tools/list/tools/call
for only this pack's tools — none of the shared meta-tools the gateway
connection above adds. Same source, same tools, no ask_pipeworx routing.
Instead of calling tools directly, you can ask questions in plain English — this works on the pack endpoint above as well as on the full gateway:
The gateway picks the right tool and fills the arguments automatically.
MIT