# US Product Recalls AI Agent Connect > US Product Recalls MCP connects your AI agent directly to the US Consumer Product Safety Commission's official database. It covers every recall published from 1973 to today. Your agent can check specific brands, models, or barcodes, browse recent safety notices, and filter by specific hazards like fire or choking risks to ensure the products you buy or gift are safe. ## Overview - **Category:** shopping - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_vN9IEUZqlOjLlceaAdGcUccRMYqs25MkjJVrb8D9/ai-agent-connect - **Tags:** product-recalls, consumer-safety, cpsc, recalls, product-safety, shopping, consumer-protection, public-data, united-states, home, baby, toys ## Description This MCP gives your AI client direct access to the US Consumer Product Safety Commission's complete record of product recalls since 1973. While the agency's own web interface often fails when users try to filter by specific hazards or UPC codes, this MCP fixes those gaps by matching your queries against the actual text of the records. It handles the messy reality of human-written data, such as when an agency describes a unit count in a long sentence or uses various ways to say no injuries were reported. Instead of getting zero results from a broken filter, your agent gets accurate, localized matches for hazards, companies, and products. You can use it to verify a single item before a purchase, track recent safety trends, or pull high-level summaries of recall activity across specific timeframes or industries. ## Tools ### check_product This tool checks a brand, model, or barcode against the CPSC database to find matching recall notices. It returns the recall number, date, hazard details, and the specific remedy the agency recommends. ### recent_recalls This tool pulls a list of all recalls published within a specific window of days. It provides a quick overview including the product, unit counts, and whether injuries occurred. ### search_recalls This tool searches the database using company names, product types, or specific hazards. It uses local text matching to ensure hazard searches like 'fire' actually find relevant notices. ### summarize_recalls This tool provides a high-level statistical view of recall activity over a set period. It breaks down the data by month, remedy type, company, and injury reports without listing every individual notice. ### get_recall This tool retrieves the full, official notice for a specific four to six digit recall number. It includes everything from manufacturer details and sale locations to consumer contact information. ## Prompt Examples **Prompt:** ``` I'm about to buy a used Char-Broil Bistro Pro electric grill from a neighbour. Has it ever been recalled? ``` **Response:** ``` I'll check the Char-Broil Bistro Pro against the database. I'll look for matches across the brand, model name, and product descriptions to see if any recalls exist for that specific grill. ``` **Prompt:** ``` What did the US recall last month? Just give me an idea of the scale, not every notice. ``` **Response:** ``` I can summarize the last 30 days for you. I'll provide the total number of recalls, how many involved injuries, the approximate number of units affected, and how they were split by company and remedy. ``` **Prompt:** ``` Are there any recalls from this year involving lead, where people actually got hurt? ``` **Response:** ``` I will search for recalls from this year that specifically mention lead and have reported injuries. I'll use local text matching to ensure I don't miss notices that describe lead exposure using different phrasing. ``` ## Capabilities ### Direct CPSC Access Your agent queries the official agency database for recalls dating back to 1973. ### Smart Hazard Filtering The MCP matches hazard keywords against the actual text of recall descriptions to avoid the broken filters on the official site. ### Injury Status Verification Your agent can distinguish between recalls with reported injuries and those where the agency explicitly stated no injuries occurred. ### Data Summarization The agent can aggregate recall counts and remedy types to show trends instead of just individual lists. ### Barcode and UPC Matching Your agent can use specific UPC codes to find exact product matches in the recall records. ## Use Cases ### Pre-purchase Safety Check Check if a secondhand item, like a baby rocker or a grill, has been recalled before you buy it. ### Trend Monitoring Ask your agent to summarize all recalls from the last month to see if a specific hazard is increasing. ### Hazard-Specific Audits Search for all recent recalls involving specific risks like choking or fire hazards. ### Detailed Recall Investigation Use a specific recall number to get the full details on manufacturers, importers, and consumer remedies. ## Benefits - Bypasses broken search filters on the official CPSC website. - Provides accurate injury reporting by interpreting varied agency phrasing. - Allows for rapid trend analysis through statistical summaries. - Enables precise product identification using brands, models, or UPCs. ## How It Works Connect your client to Vinkius to access the CPSC data instantly. 1. Connect your MCP-compatible client to the Vinkius platform. 2. Ask your agent to check a product, brand, or model. 3. The MCP queries the CPSC database, bypassing known filter errors. 4. Your agent receives the specific recall details or a summary of the data. ## Frequently Asked Questions **I searched for a product and got nothing back. Does that mean it's safe?** It means the commission has not recalled anything under that term, which is a weaker statement. This database holds recalls only, and only those the CPSC issued — products under investigation, products recalled by another agency such as the FDA or NHTSA, and products nobody has reported harm from yet are all absent. Treat an empty answer as "no recall on file" and read the hazard description on the product itself. **Why does a hazard search match words instead of categories?** Because the agency's own hazard filter is broken: it returns zero for every value asked of it, including values that appear verbatim in the data, and its hazard-type field is empty on every record. The usable text is the hazard description itself, written in prose, so the search matches your term against those descriptions — "fire" finds "fire hazard" and "catch fire", but a recall that only implies heat will not match. Use several terms if you want to be thorough. **What counts as a recall with reported injuries?** A recall whose injury field carries an actual account of harm. Roughly one recall in six phrases a no-injury outcome eight different ways — "None reported", "None reported.", "No incidents or injuries have been reported." and five others — and testing whether the field is simply non-empty counts those as injuries. This server recognises all eight and reports them as no injuries, so filtering to reported injuries keeps only recalls where the agency described harm. **Why are the unit counts described as approximate?** Because the agency publishes them as free text, not as a number. A real entry reads "About 10,605 (In addition, 342 were sold in Canada)" and another "About 7,200 (In addition, about 370 were sold in Canada)". The first figure is taken as the headline count and the full phrase is kept beside it, so you see what the agency actually wrote. Adding those first figures across a year gives the order of magnitude, not an audited total. **Do I need an API key or an account?** No. The commission publishes this database openly and the server reads it without authentication. There is no key to obtain, no quota to manage and no sign-up step.