CPG Data Analytics For AI Agents: Launching crushrewards.dev

Agentic commerce is no longer theoretical. ChatGPT's Instant Checkout has been live since September 2025 and it serves million of weekly users. Google announced its Agent Commerce Protocol (ACP) in January 2026 with Walmart, Target, and Shopify among 20+ partners. McKinsey and Google project agentic commerce will represent a $1 trillion slice of US retail by 2030, while Bain & Co estimates around $500B. Either way, it’s going to be HUGE. The “shelf” is becoming invisible and will be replaced by an agent layer where the question is no longer which product ranks on page one but which product does the agent pick for you. To pick good products, it will need to consume good data. That data today is locked behind costly paywalls and excel files. This changes today.

This week we launched crushrewards.dev — the first e-commerce pricing intelligence API built for agents, and the first CPG data service available on both x402 (Base/Solana) and MPP (Stripe/Tempo). No API keys. No contracts. Pay per query in USDC at $0.01–$0.02 a call. Your agent's wallet handles the rest.

The Invisible Shelf Is Already Here

The competitive battleground is shifting from the SEO shelf, Amazon search results and Google SERPs to the agentic shelf: whatever product the agent surfaces and purchases when a consumer says "find me the best deal." Google Cloud calls it the invisible shelf, we call it data-driven shopping.

The brands and data providers that win this decade will be the ones that understand the new buyer isn't a human scrolling a results page. It's an autonomous system making a purchase decision in under two seconds.

Agents Have Needs That the Old Data Stack Can't Deliver

Today's CPG and retail data infrastructure was built for a human workflow: quarterly category reviews, monthly dashboards, annual contracts. Agents operate on a completely different clock, and most definitely wont sift through excel files for information.

What agents need:

• Real-time prices, not weekly batches. If Walmart cuts a Tide price by 12% at 2:00pm, the agent needs to know by 2:00:30pm.
• Cross-retailer coverage in a single call. No agent is going to integrate 14 separate retailer scrapers.
• Pay-per-call billing, not annual contracts. Agents don't have procurement departments. They have wallets.
• No API keys. API keys require human onboarding, human credential management, and human-controlled rate limits. Agents can't do any of that.
• Structured, machine-first responses. Not PDFs. Not CSV exports. Not dashboards. Markdown files and APIs, baby!

What they're forced to use today in early agentic commerce products:

• Propriety data pipelines siloed behind retailers proprietary data.
• Web searches that could be stale, and lack historical context and trends.

There are no purpose built options for them. The gap is obvious the moment you try to build an agent that actually shops. The legacy providers aren't slow because they're lazy. They're slow for two main reasons: 1) because their entire business model assumes a human is waiting on the other end; 2) they have high acquisition costs for their data and cannot simply make it public of $0.02 per query.

Why x402 and MPP Matter

The plumbing here is worth understanding, because it's not a billing gimmick. It's a structural shift in how software will pay for data going forward.

x402 is Coinbase’s revival of the dormant HTTP 402 "Payment Required" status code. When an agent calls an API, the server responds with 402 and a payment specification. The agent's wallet signs a USDC micropayment — sub-two-second settlement on Base or Solana, roughly $0.0001 in transaction costs — retries the request, and gets the data. No API keys. No subscriptions. No procurement cycle. (x402.org whitepaper, Coinbase Developer Docs)

MPP (Machine Payments Protocol) is Stripe and Tempo's answer, launched March 18, 2026. It uses the same HTTP 402 primitive but is payment-method agnostic — stablecoins, cards, Lightning — and adds sessions. An agent deposits into an escrow contract once and issues vouchers for each subsequent call, so high-frequency agents don't pay chain fees on every request. MPP has been submitted to the IETF as the official HTTP 402 implementation and is already adopted by OpenAI, Anthropic, Google Gemini, Dune Analytics, and Alchemy. (mpp.dev, Stripe announcement, Fortune coverage)

Why both? They're the two de facto standards for machine-native payments, and every major agent framework is integrating both. Shipping on only one would mean cutting out half the agent ecosystem on day one. Crushrewards.dev is the first CPG and e-commerce data service on either protocol, and the first to be on both.

What You Can Expect From crushrewards.dev

Thirteen tools with API endpoints, priced at $0.01–$0.02 per query, callable from any x402- or MPP-compatible agent through your Claude Code. Payment via USDC on Solana, Base, or Tempo — the gateway auto-selects the wallet with sufficient balance.

Shopper tools ($0.01/query) — for the consumer agent making a purchase decision:

• best_price — live price comparison across retailers for a product
• price_history — daily historical pricing trend
• deal_finder — current deals in a category
• price_drop_alert — recent markdowns

Marketing tools ($0.01/query) — for the brand strategist or marketing agent:

• competitive_landscape — price positioning vs. competitors
• brand_tracker — a brand's pricing across SKUs and retailers
• promo_intelligence — current and recent promotional activity
• share_of_shelf — listing prevalence for a brand
• price_positioning — where a brand sits in the category price hierarchy

Analyst tools ($0.02/query) — for the financial or economic analyst:

• inflation_tracker — category-level inflation
• price_dispersion — cross-retailer price variance
• retailer_index — relative pricing performance per retailer
• category_summary — full category snapshot

Coverage includes Amazon, Walmart, Costco, Target, Best Buy, and other major US and Canadian retailers (we’re adding more everyday). All tools accept optional filters for country, retailer, and date range.

Sever ways to integrate:

1. MCP Server — one-line install: npx @crush-rewards/mcp-server. Drops directly into Claude Desktop, Cursor, Claude Code, or any MCP-compatible host.
2. Simply run in your Terminal npx @crush-rewards/mcp-server --setup , then send either USDC on Solana or Base, or USDC.e from Tempo to your wallet.
3. If you want free money try agentcash.dev and check out some of their cool API endpoints. From there you can ask agentcash to find our api at api.crushrewards.dev.

Even $1 gets you 50–200 queries. This is priced for agents, not procurement committees.

What this looks like in practice

The interesting thing about a pricing-intelligence API priced at one to two cents per query isn't any single use case, it's the range of people for whom that price unlocks a workflow that was previously out of reach. To make that concrete, here are four examples (based on our goals) spanning the spectrum: a regular shopper deciding when to buy, a brand manager diagnosing share loss, a buy-side analyst pressure-testing earnings guidance, and a macro economist building a real-time CPI nowcast. Same API, same tools, wildly different stakes.

What This Unlocks, by Persona

The Shopper Agent

Most price-comparison tools tell you what something costs. They don't tell you what to do. A parent staring at a $1,000 stroller doesn't need another tab open to Camelcamelcamel , they need an answer to "is now actually the right time?" Crush Rewards turns that one natural-language question into three quiet tool calls behind the scenes , best price across retailers, twelve months of price history (soon), and a check for likely upcoming drops, then hands back a recommendation, not a spreadsheet. Three cents to potentially save a hundred dollars, with no learning curve and no shopping-tool fatigue. This is the use case people will discover first, and it's the one that makes the per-query model feel obviously fair.

The Marketing Agent

Brand managers live or die by competitive pricing intelligence, but the existing tools are either expensive enterprise dashboards or spreadsheets a junior analyst rebuilds every Monday. The hard part isn't pulling the data, it's knowing which cuts to pull and building the dashboards. When a brand starts losing share, the diagnostic question ("is this a shelf-price problem, a promo problem, or a distribution problem?") usually takes days to answer because each subquestion lives in a different tool. With Crush Rewards, the agent decides the diagnostic path itself: positioning ladder, competitor promo cadence, share-of-shelf trend. Four cents, four queries, and the manager walks into the CMO's office with a real answer, and three concrete pricing plays, instead of a request for more research time.

The Financial Analyst

Alt-data is the fastest-growing line item in buy-side research budgets, and SKU-level retail pricing is one of the most valuable signals inside it. Companies like YipitData, Earnest Analytics, and Bloomberg Second Measure have built nine-figure businesses selling exactly this kind of data to hedge funds and mutual funds, but it's sold in six-figure annual contracts and consumed through static dashboards, not interrogated conversationally. Crush Rewards collapses that workflow. An analyst who's skeptical of management's "we're not discounting" line on the last earnings call can fact-check it at the SKU level in five minutes for two cents, build a falsifiable thesis with specific evidence (hero SKU promo days up 7.5x year-over-year, first sub-$80 print since 2021), and walk into the morning meeting with a position recommendation. Institutional-grade alt-data on a per-query basis is a genuinely new unit economic.

The Macro Researcher

Real-time inflation measurement is one of the longest-running open problems in applied macro. Official CPI prints lag by four to six weeks and use baskets that don't decompose cleanly by trade exposure or geography, which is exactly why MIT's Billion Prices Project (now sold commercially as PriceStats by State Street) was founded in the first place. Central banks pay institutional rates to access it. Crush Rewards puts the same primitive in the hands of any economist with an MCP-compatible agent. A researcher studying FX pass-through can build a real-time nowcast in four minutes: same-SKU dispersion across the US-Canada border isolates pure exchange-rate effects, category-level decomposition reveals which goods retailers pass through fast versus slow, and the whole study costs ten cents. For an asset class where being two weeks ahead of the next CPI print is worth millions, that isn't really a price — it's a free option.

Why We Can Serve This Data

Today, the pricing feed behind crushrewards.dev comes from a comprehensive e-commerce ingestion layer: real-time scraping and normalization across Amazon, Walmart, Costco, Target, Best Buy, and the other major US and Canadian retailers, collapsed into a single schema so an agent can compare across retailers in one call. That's what's powering every query you run today.

But the roadmap is what makes this durable. Crush is a consumer app first. Users are earning CRUSH tokens for contributing their own pricing data, snapping photos of store shelves (soon), submitting receipts, and linking their debit and credit cards. We’ve already distributed over 45K CRUSH tokens. As that contributor base grows, that first-party, permissioned, human-verified data flows directly into the same API. E-commerce scrape today; hybrid scrape + crowdsourced shelf and receipt data tomorrow.

The comparison is stark:

• Legacy CPG data firms: panel-based, extrapolated, weekly-to-monthly latency, and sometimes can costs hundreds of thousands.
• Keepa and single-retailer scrapers: Amazon-only, fragile, no cross-retailer normalization.
• Crush today: cross-retailer e-commerce ingestion, normalized to a single schema, real-time, pay-per-query.
• Crush tomorrow: the above plus a first-party network of consumers paid in CRUSH tokens to contribute receipts, card-linked transactions, and in-store shelf photos, the only data source covering prices the scrapers can't see (in-store tags, regional markdowns, club-store SKUs).

Consumers earn tokens. Agents query data. Brands and analysts get an answer for a penny. The value loop that was broken for twenty years finally closes. For a deeper look at how Crush Rewards v2 expanded the contributor side of this network, that post covers the app changes that make this data layer possible.

Go Try It Right Now

If you have Claude Code, open it and type:

npx @crush-rewards/mcp-server --setup

No signup. No API key. No credit card. Your agent pays a penny in USDC, Crush returns live prices, and you just ran the first agentic commerce query most marketers haven't even drawn on a whiteboard yet. Your just going to need to send $1-$2 USDC from your Coinbase, or whatever crypto platform you use.

Developers: npx @crush-rewards/mcp-server — live in 30 seconds.

Marketers: Book a 20-minute demo by sending us a message at azad@crushrewards.app

Agent builders: We're the first CPG data source on x402 and MPP. Ship your agent on top.

The agents are already shopping. The question is whether you're feeding them.