For AI agents and developers
Spoon Scout for agents
Spoon Scout reports what labs, regulators and recalls have found in US packaged foods, with the source and date on every number. This address serves the same data to AI assistants and code: an MCP server and a JSON API. No key, no sign-up.
Data as of 2026-11-16: 445,968 products, 2,723 with lab results, 29,598 FDA recalls.
Looking for the website? People should use https://spoonscout.com/.
Connect
MCP server (Streamable HTTP, no authentication, read-only tools):
https://food.agentlookups.ai/mcp
- Claude Code: run
claude mcp add --transport http spoon-scout https://food.agentlookups.ai/mcp - Claude, ChatGPT and other assistants: add a custom connector with this URL and leave authentication empty. Menu names differ by app and plan.
Steps for each assistant: Using these with Claude, ChatGPT and other assistants.
MCP tools
- Check a packaged food
check_product - Find where a chemical comes from in food
chemical_sources - Compare packaged foods
compare_products - Typical contaminant levels for a kind of food
food_category - Check a food additive's legal status
additive_status - Compare a test result with the US population
interpret_test_result
JSON API
Plain GET requests, no key, CORS open. Every answer carries its sources, the notes below and the terms. The same list as JSON: https://food.agentlookups.ai/v1.
/v1/search?q=<product, brand, barcode, chemical or additive>[&page=<n>][&limit=<n>][&additive=<slug>]Matching products (lab-tested first, package sizes folded), chemicals, additives and kinds of food. A chemical plus a food (q=lead in cinnamon) adds that chemical's lab results for the food and FDA's typical level; additive=<slug> keeps products whose ingredients list it. Try it: /v1/search?q=milk/v1/product?barcode=<UPC/EAN/GTIN> | id=<id> | q=<name>One product: lab results by chemical, FDA recalls naming its barcode, restricted additives, and FDA's typical levels for its kind of food; photo, when one passed our checks, is a front-of-pack photo from Open Food Facts with the credit and licence (CC BY-SA 3.0) to show with it. A name that fits several products returns candidates; call again with id. Try it: /v1/product?barcode=00000004978903/v1/chemical/{slug}?[kind=<category slug>]Foods where FDA found the highest typical levels (highest_typical_foods, ranked by one serving's share of a toddler's daily limit when the chemical has one), the 10 named products with the highest lab results (kind narrows them to one kind of food; tested_kinds lists the kinds), recent recalls, and a drinking-water note. Slugs include lead, perchlorate, arsenic-inorganic, dehp and bpa. Try it: /v1/chemical/perchlorate/v1/category/{slug}?[sort=<chemical slug>][&order=high][&all=1]FDA's typical levels for a kind of food, how close FDA's foods are as a stand-in, and the products of that kind with lab results: the 50 lowest first in the most-tested chemical by default (below the lab's limit first, then measured amounts), with tested_total giving the full count. sort ranks by another chemical, order=high puts the highest first, all=1 returns every tested product. Slugs include milk, rice and baby-infant-foods-beverages. Try it: /v1/category/milk/v1/additive/{slug}Where an additive is allowed, restricted or banned (FDA, EU, US states), each with its law and date, and how many products list it. Try it: /v1/additive/titanium-dioxide/v1/compare?ids=<id>,<id> (2 to 5 product ids)Products side by side, chemical by chemical: each product's own lab results where they exist, else the typical level for its kind, marked as typical. Try it: /v1/compare?ids=splasticlist-100,splasticlist-121/v1/result?chemical=<slug>[:<NHANES code>]&value=<number>[&unit=ug/L|ug/dL|ug/g creatinine][&age=all|young|6-11|12-19|20+|<years>]Where a urine or blood test result falls among US population percentiles (CDC NHANES) for everyone or one age group, per liter or per gram of creatinine: a tier in words (far_above at twice the 95th percentile, with a line telling the person to see their clinician), a unit_check when the value looks like a unit mix-up, CDC's and NIOSH's blood lead guidance for children and adults, and cited context. Context, not a diagnosis. Try it: /v1/result?chemical=perchlorate&value=5/v1/sourcesEvery source with the publisher's date, our fetch date, record count and license. Try it: /v1/sources
How to read results
- Every number names its source and date. 'Typical' values are FDA averages for a kind of food, not tests of the product.
- 'Not tested' means no published test was found, not that the food is free of the chemical.
- A result below a lab's limit was not reliably measurable; it is not a zero.
- Maker-disclosed California AB 899 results come from a compiled dataset that leaves out many lots where nothing was detected, so they can look worse than the maker's full record.
- Lab results attached to a barcoded product by brand and name (name_matched_lab_records) are likely, not certain, matches: the lab recorded no barcode.
- This is a record of what was published, not medical or nutrition advice.
Terms and limits
Free to use. Lab results are CC-BY-SA 4.0 (detectlead.com requires share-alike); credit each original source named in the record. Government data is public domain. Product photos are CC BY-SA 3.0 from Open Food Facts: show each with the credit and licence in its photo fields.
People and AI agents are welcome. Please keep it under a few requests a second.
More
- llms.txtThis page as plain text for agents.
- /v1The endpoint list as JSON.
- Sources and licensesEvery publisher, its date and its license.
- Spot a mistake? Email corrections@agentlookups.ai with the address and what you believe is right.