The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Jecfa listing page.
Safety evaluations from the Joint FAO/WHO Expert Committee on Food Additives: Acceptable Daily Intakes (ADIs), the committee's conclusions, dietary-exposure findings, and the meeting reports and toxicological monographs behind them — for food additives, flavourings, food contaminants and veterinary drug residues.
Part of Pipeworx — an MCP gateway connecting AI agents to 1683+ live data sources.
JECFA is the body whose ADI a regulator cites when it sets a limit. This pack answers "has this substance been evaluated, and what intake did the committee judge acceptable" — not "is it permitted in yogurt in the EU", which is the Codex GSFA / national-regulator question.
| Tool | What it answers |
|---|---|
jecfa_search | Look up by substance name, CAS number, or INS/E-number → ADI, CAS, functional class, chemical id |
jecfa_chemical | Full record for one id: synonyms, INS, and every evaluation JECFA has made — year, ADI, conclusions, intake findings, meeting number, TRS report, tox monograph |
jecfa_by_functional_class | Every substance in a class (sweetener, preservative, colour…) or in one of the four groups |
jecfa_functional_classes | The 71-class vocabulary, so a caller knows the exact name to pass |
jecfa_by_number | Resolve a FEMA or JECFA substance number |
jecfa_recent_evaluations | What the committee has looked at most recently |
3,245 substances: 777 food additives, 2,289 flavouring agents, 115 veterinary drugs, 64 food contaminants. Evaluations run from the committee's earliest meetings to the present — aspartame, for instance, carries evaluations from 1981 through the 2023 re-evaluation at meeting 96.
None. Keyless, no registration, no quota.
The database has no documented API. The endpoints this pack calls are the ones backing the site's own Kendo grid, read off the page. They are undocumented and unversioned, so they can drift without notice; the pack's own gotcha comments (top of src/index.ts) record the shapes that were verified live.
Five upstream behaviours produce confidently wrong answers if taken at face value. All are guarded in the pack, but they matter if you extend it:
/SearchChemical/ByPartialName fabricates a row. It prepends your query string as a result with Id: 0 and a blank ADI — the UI's "search for this literal text" affordance. Pass it through and you have invented a substance that was never evaluated. The pack drops every row with Id <= 0.
The same endpoint returns CAS_NO: null even when you searched by CAS. Searching 50-70-4 correctly finds SORBITOL, then reports its CAS as null. Only /ChemicalData/GetBy/par/ returns populated records, which is what jecfa_search actually calls.
/ChemicalData/GetBy/fir/ wants the character, not the id. Its companion /FilterData/Get/FirstCharacter hands you {"Id": 65, "Name": "A"}; passing 65 returns HTTP 500, passing A returns 501 rows.
INS/E-numbers are not searchable at all. ByPartialName/951 returns zero rows — INS numbers exist only on the server-rendered detail page. So jecfa_search resolves them through INS_INDEX, a build-time harvest of all 777 additive detail pages (486 carry an INS). Regenerate with node mcps/jecfa/scripts/build-ins-index.mjs after a meeting adds substances.
Detail-page citation rows use a different CSS class than every other field. ADI, comments, intake and meeting render in col-sm-8; the Report, Tox Monograph and Specification rows render in col-**md**-8 and wrap their text in an <a>. A parser matching only col-sm-8 returns null for all three — which reads as "JECFA published no report for this evaluation" when in fact the link to the TRS PDF was right there. The url on each citation is upstream's own, usually a direct PDF on iris.who.int.
Bare INS numbers are genuinely ambiguous — INS 160 denotes 19 separately evaluated carotenes — so the index maps each key to a list and jecfa_search returns every candidate with ambiguous: true rather than picking one. Upstream writes sub-forms inconsistently (100(i) and 100i both occur), so keys are normalized to alphanumerics; the generator and the pack must keep using the same normalization or every lookup silently misses.
Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):
tools/list at https://gateway.pipeworx.io/jecfa/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 1683+ 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/jecfa_search. 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