Veterinary FDA — Approved Animal Drugs, Adverse Events, Pet Product Recalls, Pet Services

Approved animal drugs, side-effect reports, product recalls, and nearby pet services for the animal/veterinary side of the world — dogs, cats, horses, cattle. Three sources: the FDA Green Book (Animal Drugs @ FDA — the official list of approved new animal drug applications, keyless), openFDA's Center for Veterinary Medicine adverse-event database (985,705 Dog / 148,091 Cat reports as of 2026-09-04, keyless) and OpenStreetMap (vets, dog parks, pet stores, shelters). No pack in the MCP ecosystem serves the animal adverse-event data before this one — openfda never touches it, and openfda's drug endpoints are human-drug only, so the Green Book side is likewise unserved.

Part of Pipeworx — an MCP gateway connecting AI agents to 1684+ live data sources.

Together these answer one question sequence end to end: what is approved for this animal (Green Book) → what has it done to animals (adverse events) → has it been recalled (enforcement reports).

Why this matters for AI agents

"Is this side effect from my dog's flea medication normal?" and "has this dog food been recalled?" are questions people actually ask, and until now nothing answered them from primary-source FDA data.

Common flows:

• What is approved. "What FDA-approved drugs contain carprofen?" → animal_drug_search({query: "carprofen"}) → 24 approved applications with sponsor, status and approval date. "What is approved for dogs?" → animal_drug_search({species: "Dogs"}) → 694.
• What an approval actually covers. "What was Rimadyl approved for?" → animal_drug_search({brand_name: "Rimadyl"}) → applicationId 1024 → animal_drug_detail({application_id: 1024}) → FDA's own FOI approval summaries, per-species dosage and indications, and the labelling PDFs.
• Side effects for a drug. "What side effects have been reported for Bravecto in dogs?" → vet_adverse_events({drug: "Bravecto", species: "Dog"}) → individual reports with reactions, outcome, dose.
• Aggregate safety picture. "How common are seizures with fluralaner in dogs?" → vet_adverse_event_summary({active_ingredient: "fluralaner", species: "Dog"}) → counts by reaction, outcome, year.
• Recalls. "Has Blue Buffalo dog food been recalled?" → vet_product_recalls({product: "Blue Buffalo"}).
• Nearby services. "Emergency vet near Austin TX" → pet_services_near({place: "Austin TX", kind: "emergency_vet"}).

Auth

Public, keyless throughout.

The Green Book tools (animal_drug_*) call animaldrugsatfda.fda.gov directly — the Animal Drugs @ FDA SPA's own backing API. It is a different upstream from openFDA: no key applies to it, PLATFORM_DATAGOV_KEY is not sent, and _apiKey does nothing there.

api.fda.gov runs on api-umbrella (the same api.data.gov stack as openfda), so the gateway injects PLATFORM_DATAGOV_KEY (workers/gateway/src/index.ts, MCP_PACKS entry for veterinary-fda) to raise the anonymous 40 req/min · 1,000/day cap to 120,000/day — the same platform key already backing 15 other federal packs (openfda, fac, nasa, govinfo, regulations-gov, etc.). Bring your own key with _apiKey to draw on your own quota instead. A rejected/revoked platform key falls back to anonymous rather than failing the call.

Nominatim (geocoding) and Overpass (pet_services_near) are separately keyless, public OSM infrastructure.

Tools

Data traps (measured 2026-09-04, load-bearing for how these tools are built)

• drug.brand_name is near-useless. In the adverse-event endpoint it almost always holds the registrant's internal short code (e.g. "MSK" for Merck Animal Health), not the marketed product name. Searching drug.brand_name for "Bravecto", "Frontline", "Heartgard", or "Simparica" returns zero matches for all four. The actual product name is recorded inside drug.active_ingredients.name instead (mixed in with real chemical ingredient names — messy upstream data). vet_adverse_events/vet_adverse_event_summary try brand_name first, then fall back to active_ingredients.name automatically and report resolved_from so the caller knows which field actually matched.
• count=<field>.exact is per-field, not universal. count=reaction.veddra_term_name.exact and count=outcome.medical_status.exact both work; count=animal.species.exact and count=original_receive_date.exact both return "Nothing to count". vet_adverse_event_summary's by-year table works around the date-field limitation by requesting count=original_receive_date (which buckets per exact DAY, not year) and aggregating client-side.
• Bare multi-word terms are an OR, not an AND, and openFDA does not rank by relevance. search=product_description:dog+food (no quotes) matches ~29,000 records — mostly hot dogs and corn dogs, because "dog" OR "food" hits practically the whole food-recall corpus, returned in whatever order openFDA feels like. Only the phrase-quoted form, search=product_description:"dog food", returns the real 2 pet-food matches (Freshpet). vet_product_recalls always phrase-quotes multi-word product terms for this reason, and never falls back to a loose OR match — for a recall/safety answer, a false-positive (a "Blue Bell" ice-cream recall surfacing for a "Blue Buffalo" query) is worse than an honest zero, so unlike this pack's own adverse-event fallback and unlike openfda's free-text fallback, there is no loose-match tier here.
• Pet-food recall coverage in openFDA is genuinely incomplete, not a search-syntax problem. There is no dedicated animal/pet-food recall endpoint — vet_product_recalls reuses food/enforcement.json and drug/enforcement.json, the same feeds openfda uses for human food/drugs. Several well-known, FDA-confirmed pet-food recalls are absent from this dataset entirely, verified by exhausting every plausible product/firm-name spelling (including the exact names from FDA's own press releases) rather than by a single failed query: Blue Buffalo's 2010/2017/2022 recalls and the 2021 Sportmix aflatoxin recall (~70 dog deaths, a formal FDA Class I recall) all return zero records under any search. FDA's own recall-history page (fda.gov/animal-veterinary/safety-health/recalls-withdrawals) is not the same backing data as openFDA's Enforcement Reports API, and openFDA does not mirror it. vet_product_recalls returns total: 0 plus an explicit note in that case — treat a zero as "no Enforcement Report record found," never as "never recalled."

Green Book traps (measured 2026-09-05)