codex-contaminants

Codex Alimentarius CXS 193-1995 maximum levels for contaminants and toxins in food — lead, cadmium, arsenic, mercury, methylmercury, tin, aflatoxins, ochratoxin A, patulin, melamine, hydrocyanic acid — by commodity.

Part of Pipeworx — an MCP gateway connecting AI agents to 1576+ live data sources.

59 maximum levels, 2015 edition. Both of those numbers are limitations, and both are stated on every response rather than left in this file.

Read this before using a value

Edition. Codex publishes CXS 193 only as a document, and the current text is behind an authenticated FAO workspace. This pack carries the newest public official edition, 2015 (Amended in 2010, 2012, 2013, 2014, 2015). The standard has been amended in 2016, 2017, 2018, 2019, 2021, 2022, 2023, and at CAC47 (Nov 2024) and CAC48 (Nov 2025). Those amendments added maximum levels — lead in cinnamon (2.5 mg/kg) and dried culinary herbs (2.0 mg/kg) among them.

Coverage. 59 entries from the Schedule I tables. Rows whose commodity name wraps across table lines were dropped rather than risk labelling a legal limit with a sentence fragment like "exception of berries".

Together: a miss is not a finding that no limit exists. The not-found path says so in words, because "no maximum level" reads as permission.

Every value carries the source PDF page, so any figure can be checked in one step against https://www.fao.org/input/download/standards/17/CXS_193e_2015.pdf.

Related MCP server: food-safety-mcp

Tools

• contaminants_max_levels({contaminant, commodity?}) — Codex MLs for a contaminant, optionally narrowed to a food.

• contaminants_for_commodity({commodity}) — every contaminant limit that applies to one food.

Extraction, and what went wrong four times

extract.py is the extractor. It is column-band driven for a reason: each of these was a real, silent failure during the build, and the tests pin all of them.

1. Character offsets do not work. The ML column sits at character 36 on one page and 29 on the next, so fixed slicing attaches a real limit to a truncated commodity name.

2. Reading any number on the page swept in, under LEAD, 73 (a JECFA meeting number), 25 (the withdrawn PTWI in µg/kg body weight) and 3 (3 mmHg of blood pressure) as maximum levels. Fixed by taking the ML column's x-band from each page's table header and accepting only numbers inside it.

3. Units differ per contaminant — aflatoxins are µg/kg, metals mg/kg. The unit is read from the header too. Hardcoding mg/kg made aflatoxin M1 in milk read 0.5 mg/kg instead of 0.5 µg/kg: wrong by 1000×.

4. Section headings must come from the document, not memory. The standard says DEOXYNIVALENOL (DON), and a 3-letter heading TIN is easy to filter out by accident — when it was, tin's canned-meat limits (50 mg/kg) were filed under methylmercury.

To refresh: point extract.py at a newer PDF, re-run, and read the output before shipping it. The tables are small enough that a human check pass is feasible, and for legal limits it is not optional.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "codex-contaminants": {
      "url": "https://gateway.pipeworx.io/codex-contaminants/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/codex-contaminants/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:

{
  "mcpServers": {
    "pipeworx": {
      "url": "https://gateway.pipeworx.io/mcp"
    }
  }
}

Both URLs reach the same gateway and the same 1576+ data sources. The only difference is which pack's tools are listed directly; ask_pipeworx reaches all of them from either one.

Standalone (no gateway account)

This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:

{
  "mcpServers": {
    "codex-contaminants": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-codex-contaminants"]
    }
  }
}

Or run it directly to confirm it starts:

npx -y @pipeworx/mcp-codex-contaminants

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.

Using with ask_pipeworx

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:

ask_pipeworx({ question: "your question about Codex Contaminants data" })

The gateway picks the right tool and fills the arguments automatically.

More

• Docs and guides

• pipeworx.io

License

MIT