Built for AI agents

Your agent is good at reasoning. It should not be guessing at food law. fmcg.network gives it a set of narrow, read-only checks that return facts plus the source behind them, so every answer can be traced and verified.

Connect in one line

https://mcp.fmcg.network/mcp

Works with any MCP client: ChatGPT connectors, Claude, VS Code and custom agents built on MCP SDKs. Read-only checks run without an account (20 per hour, 60 per day per network); a free sign-in lifts the limit.

How an agent uses it

  1. search_checks with a plain-language need.
  2. describe_check to read inputs, outputs and scope.
  3. run_check with structured inputs.
  4. get_source to attach the citation to the answer.

The tools

search_checks
Find checks by plain-language query and market (us, eu, uk), for example "FSMA 204 receiving KDEs" or "undeclared sesame recall". Returns check_ids.
describe_check
Read a check's inputs, output, jurisdiction and coverage, plus a ready run_check call.
run_check
Run one check by check_id with structured arguments and get the result with its sources.
get_source
The registers, regulations and references a check answers from, with licence and citation text.
list_regulations
The regulations and live registers covered, each mapped to its check_ids.
lookup_nutrients
Fast path to USDA FoodData Central: best match for a food and its nutrients per 100 g, with the record's data type and date.

Clients that call checks by name can still get the full list: add ?tools=full to the endpoint URL or passfull: true to tools/list.

What you can rely on

  • Cited by default. Results name their source: FDA enforcement reports, USDA FoodData Central, UK Food Standards Agency alerts, and the eCFR and EUR-Lex provisions a rule check reads.
  • Says what it did not check. Every answer carries the dataset's coverage, and a partial dataset says that no match is not an all clear.
  • Narrow and predictable. One check, one question, structured output. Easy to chain inside agent workflows.
  • Read-only. Checks never write to your systems.
  • US and EU coverage. Nutrition, recalls, traceability, importer verification, labelling, allergens, additives and claims.

Typical workflows

  • QA agent: screen a new SKU's ingredient list against the EU allergen list and recent FDA recalls.
  • Regulatory agent: build the EU nutrition declaration under Regulation (EU) No 1169/2011 before an EU launch.
  • Procurement agent: list the FSMA 204 Key Data Elements and the FSVP file gaps for a new foreign supplier.
  • Product development agent: pull USDA FoodData Central nutrient data to draft a Nutrition Facts panel.

Example call

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "run_check",
    "arguments": {
      "check_id": "recall_watch__search_us_recalls",
      "arguments": {
        "query": "undeclared sesame"
      }
    }
  }
}

Cited answers to start from

Frequently Asked Questions

Do I need an account to use fmcg.network from an agent?
No. Read-only checks run without signing in, limited to 20 checks per hour and 60 per day per network. Signing in with a free account from your AI client lifts that limit.
What is the fmcg.network MCP endpoint?
https://mcp.fmcg.network/mcp. It speaks the Model Context Protocol over HTTP, so any MCP client can connect to it.
Can my agent still call individual checks by name?
Yes. tools/list returns six entry points by default to keep context small. Add ?tools=full to the endpoint URL, or pass full: true to tools/list, to get every check as its own tool.
Does fmcg.network give legal advice?
No. It returns facts and the sources behind them, says what a dataset does not cover, and leaves the decision to your team.

Not legal advice. fmcg.network returns facts and citations. Your team makes the call.