Pet Recall Watcher MCP

An open-source Model Context Protocol server that monitors pet food and veterinary drug recalls from the Government of Canada (CFIA / Health Canada) and the US FDA (openFDA), making recall data queryable by AI assistants like Claude.

Built by AgentProLabs · Open source under MIT

Status: v0.1. The npx package is pending its first npm publish; the landing page is live at recalls.agentprolabs.com. A hosted HTTP/SSE MCP endpoint is planned for v0.2 — for now use the stdio server below.

What it does

Recall communication at the vet-clinic level is a scramble of distributor emails and listservs, with no machine-readable feed. Pet Recall Watcher ingests the public government recall datasets daily and exposes them over MCP so an AI assistant can answer questions on a clinic's behalf.

• Daily ingestion of pet-relevant recalls from the Government of Canada open dataset and the openFDA food + drug enforcement endpoints.
• Three MCP tools:
  • query_active_recalls — search/filter active recalls by species, category, region, and date range.
  • lookup_product_status — fuzzy-match a product name against the recall database (handles partial names and minor misspellings).
  • subscribe_recall_alerts — register an email for filtered recall alerts.
• Email notifications via Resend when new recalls match a subscription.
• Landing page with a subscribe form and a recent-recalls feed.

It runs as a local stdio server (via npx) for desktop AI clients.

Quick Start (stdio)

Run directly with npx — no install needed:

npx @agentprolabs/pet-recall-mcp

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "pet-recall-watcher": {
      "command": "npx",
      "args": ["-y", "@agentprolabs/pet-recall-mcp"]
    }
  }
}

Restart Claude Desktop — the tools appear automatically. Then ask, e.g., "Are there any recent dog food recalls in Canada?"

Requirements: Node.js 20+. No configuration needed — the server is a thin client over the hosted Pet Recall API, so there are no API keys or database credentials to set. (Advanced/self-host: override the API origin with the PET_RECALL_API_URL env var.)

Tools

query_active_recalls

Returns active pet food and drug recalls, with optional filtering. Sorted by recall date (newest first), paginated.

| Parameter | Type | Default | Description | | ------------------ | -------------------------------------------------------------- | ------- | ---------------------------------- | | species | "dog" | "cat" | "all" | "all" | Filter by affected species | | product_category | "food" | "treat" | "drug" | "supplement" | "all" | "all" | Filter by product type | | region | "canada" | "us" | "all" | "all" | Filter by recall origin | | since_date | ISO 8601 date string | — | Only recalls on or after this date | | limit | number (1–100) | 20 | Results per page | | offset | number | 0 | Pagination offset |

Example response:

{
  "recalls": [
    {
      "id": "1e6a57aa-80f6-400a-9896-a7f90baf4cfc",
      "title": "Primal Patties for Dogs Beef Formula recalled due to Listeria",
      "source": "canada_recalls",
      "source_id": "64441",
      "species": ["dog"],
      "product_category": "food",
      "product_names": ["Primal Patties for Dogs – Beef Formula"],
      "brand": null,
      "risk_level": null,
      "reason": "Microbial hazard",
      "region": "canada",
      "recall_date": "2022-08-05",
      "source_url": "https://recalls-rappels.canada.ca/en/alert-recall/...",
      "data_freshness": "2026-06-02 23:17:23+00"
    }
  ],
  "total_count": 1,
  "limit": 20,
  "offset": 0
}

lookup_product_status

Check whether a specific product has been recalled. Uses PostgreSQL pg_trgm fuzzy matching against product names and brand — partial names and minor misspellings still find results. No match is a valid, useful answer (not an error).

| Parameter | Type | Required | Description | | ----------------- | ----------------- | ---------- | ----------------------------------------------- | | query | string (≥2 chars) | Yes | Product name, brand, or UPC to look up | | match_threshold | number (0.1–1) | No (0.3) | Minimum similarity score (lower = more results) |

Example response (match found):

{
  "matches": [
    {
      "title": "FRESHPET SELECT ... REFRIGERATED DOG FOOD",
      "source": "openfda_food",
      "risk_level": "Class II",
      "recall_date": "2019-07-31",
      "source_url": null,
      "similarity_score": 0.72,
      "matched_field": "product_name"
    }
  ],
  "query": "freshpet dog food",
  "total_matches": 1,
  "data_freshness": "2026-06-02 23:17:23+00"
}

Example response (no match):

{
  "matches": [],
  "query": "asdfghjkl",
  "total_matches": 0,
  "data_freshness": null
}

subscribe_recall_alerts

Subscribe an email to recall alerts matching optional filters. A duplicate (same email + filters) returns the existing subscription rather than creating a new one.

| Parameter | Type | Required | Description | | ------------------ | -------------------------------------------------------------- | ------------ | ---------------------------------------------------------------- | | email | string | Yes | Email address for alerts | | species | "dog" | "cat" | "all" | No ("all") | Only alert for this species | | product_category | "food" | "treat" | "drug" | "supplement" | "all" | No ("all") | Only alert for this category | | region | "canada" | "us" | "all" | No ("all") | Only alert for this region | | webhook_url | string | No | RESERVED for Phase 2 PMS integration — accepted but not acted on |

Example response:

{
  "subscription_id": "e5f6g7h8-...",
  "email": "[email protected]",
  "filters": {
    "species": "dog",
    "product_category": "food",
    "region": "canada"
  },
  "status": "active",
  "created_at": "2026-06-02T00:00:00Z",
  "message": "Subscription created"
}

Resources

recent-recalls

URI: recall-feed://recent

Returns the 20 most recent recalls across all sources as JSON — lets a client show current recall activity without calling a tool.

Prompts

check-product-safety

A pre-built prompt that takes a product name, runs the fuzzy lookup, and frames the result for a plain-language safety summary.

| Argument | Type | Required | Description | | -------------- | ------ | -------- | -------------------- | | product_name | string | Yes | The product to check |

Server Info

| Field | Value | | --------- | -------------------------------------- | | Name | pet-recall-watcher | | Version | 0.1.0 | | Protocol | MCP (Model Context Protocol) | | Transport | stdio (local); HTTP/SSE planned (v0.2) | | Tools | 3 | | Resources | 1 | | Prompts | 1 |

Data sources

Daily ingestion from three public, unauthenticated government datasets — CFIA (Canada), openFDA food, and openFDA drug enforcement. See docs/DATA-SOURCES.md for endpoints, schema mappings, freshness caveats, and the pet-relevance filtering strategy.

Development

npm install
npm run typecheck   # tsc --noEmit
npm run lint        # eslint .
npm run test        # vitest run
npm run build       # tsc -> dist/

Architecture: shared logic in src/ (data-sources, tools, db, notifications); Cloudflare Workers in workers/ (mcp-server, ingestion, api, notification); the Astro landing page in landing/. See IMPLEMENTATION_LOG.md for the build playbook and docs/PHASE2-HOOKS.md for deferred features.

Coming later

• Hosted HTTP/SSE MCP endpoint at recalls.agentprolabs.com/mcp (v0.2).
• match_recall_to_patients — cross-reference a recall against a clinic's patient product records (Phase 2; see docs/PHASE2-HOOKS.md).

Built by AgentProLabs · Source · Open source under MIT.