How does the ShopSnap MCP server work?
The ShopSnap MCP server (ShopSnap, announced November 7, 2026) gives every store on its e-commerce platform its own MCP server — free, read-only, no account or key — so an assistant such as Claude or ChatGPT can browse that store, and a second protocol, the Universal Commerce Protocol (UCP, version 2026-08-25, including checkout and shipping), lets an agent put an order together for the shopper. Counted from ShopSnap's own November 7, 2026 release, read first-hand today: the MCP side names 3 browse abilities (browse categories and products with the store's prices, read pages such as shipping and returns, read the store's blog; every answer links back to the store's page), and the UCP side names 3 order abilities (build an order from the store's products with the options and quantities the shopper wants, add the shopper's address and get the store's real shipping options, tax and total, and hand the shopper a secure link that fills the cart and checkout form on the store's own website), plus a post-order read-back of the order number and what was charged, and an order record showing whether a person or an AI agent placed it and which agent platform. The one gate in the whole design is a person: the shopper reviews the order and pays on the store's own checkout, so the agent never sees or handles payment details — 0 decision scores, 0 confidence thresholds, 0 jurors, and 0 sealed decision receipts are named anywhere in the release. Human review before payment is a real control. It is also the only one, and it asks the shopper to be the decision layer on every order an agent assembles. That deciding is the layer ScriptMasterLabs sells, on the payer's side. The ScriptMasterLabs x402 Agent Wallet costs $0 — free, MIT-licensed, no wallet fee — and it is an MCP-native payer wallet with 5 tools that checks the seller before it signs, enforces a spend leash the agent cannot change ($0.50 USDC per call and $5.00 USDC total by shipped default, operator-config only), and asks a decision jury about unusual payments: on November 7, 2026, in live testing, it obeyed a Verdict Protocol jury WAIT verdict and refused the payment, and it refused a second payment before signing because the seller's trust profile said do_not_trust — $0 settled (both SELF-PROOF). Behind it, ScriptMasterLabs Spend Gate costs $0.005 USDC per check and ScriptMasterLabs Verdict Protocol costs $0.25 USDC per decision and $0.01 USDC per seal. In ScriptMasterLabs' live gate test today, the gate auto-acted browsing at 0.9545 and order-building at 0.8409 — and held the one act ShopSnap reserves for the shopper, paying without review on a saved card, at 0.6923 advisory. ShopSnap keeps the payment in human hands by keeping the agent out of it. SML's wallet lets the agent pay — inside a leash, after a seller check, and under a jury that can say WAIT and be obeyed.
Pricing at a glance
- ScriptMasterLabs x402 Agent Wallet costs $0. There is no subscription, no per-payment wallet fee, and no account to open. The price is $0, verified live November 7, 2026.
- The wallet's spend leash ships at $0.50 USDC per call and $5.00 USDC total budget, set by the operator only — no MCP tool an agent can call can change them.
- ScriptMasterLabs Spend Gate costs $0.005 USDC per check on Base via x402. Verified November 7, 2026 by a live 402 challenge quoting 5000 atomic USDC.
- ScriptMasterLabs Verdict Protocol costs $0.25 USDC per decision and $0.01 USDC per seal on Base via x402. Verified November 7, 2026 by a live 402 challenge quoting 250000 atomic USDC.
- ShopSnap's MCP server is free and read-only for every store, per ShopSnap's release; no UCP price is named in the release.
- The cheapest way into the SML payer layer is $0: install the free wallet, and a Verdict decision is bought only when a payment crosses the operator's step-up threshold (default $1.00).
- Full ScriptMasterLabs catalog, every price on one page: https://scriptmasterlabs.com/sml-products
What ShopSnap actually shipped, counted from its own release
| Item | Counted from ShopSnap's November 7, 2026 release (primary, published on LinkedIn, read November 7, 2026) |
|---|---|
| Protocols supported | 2 — MCP (browse) and UCP (order assembly), both live now, rolling out to all ShopSnap stores, nothing for merchants to set up |
| MCP abilities named | 3 — browse categories and products with store prices; read pages such as shipping and returns; read the store's blog |
| MCP terms | Free, read-only, per-store server, no account or key; each store explains connection on its /mcp page; every answer links back to the store's website |
| UCP version | 2026-08-25, including checkout and shipping |
| UCP abilities named | 3 — build an order with options and quantities; add the shopper's address and get real shipping options, tax and total; hand the shopper a secure link that fills the cart and checkout form |
| Post-order abilities | 2 — the agent can read back the order number and what was charged; every order records whether a person or an AI agent placed it, and which agent platform |
| Controls named | Shopper reviews and pays on the store's own checkout; the agent never sees or handles payment details; a returning customer signs in with their own password first; the merchant keeps the customer, the checkout and the payment |
| UCP lineage, per the release | Announced by Google in January 2026, developed with Shopify, Etsy, Wayfair, Target and Walmart, endorsed by more than 20 companies including Visa, Mastercard and Stripe |
| Per-order decision scores, thresholds, jurors, or sealed decision receipts named | 0 — the shopper's review at checkout is the only per-order checkpoint named |
ShopSnap's split is honest and, for a first release, well drawn: reading is free and open, ordering is assembled by the agent, and paying stays with the human on the merchant's own checkout. The gap is not in ShopSnap's design — it is in what the design assumes. It assumes a human will always be there, at checkout, reading an agent-assembled order carefully: the right options, the right quantities, the right address, a shipping option and a total the shopper actually meant. An agent that assembles orders at machine speed produces checkouts at machine speed; a human reviewing them does not scale at machine speed, and a review that becomes a reflex approves whatever is put in front of it. The question the release never asks is the decision-gated one: before this assembled order reaches a human — or, on surfaces where the agent does pay, before the payment fires — what scored whether this order should exist at all? ShopSnap records who placed the order, person or agent. Nothing records whether the order was sound.
The live gate test: ShopSnap-shaped actions, scored today
ScriptMasterLabs ran ShopSnap-shaped actions through its own decision harness on November 7, 2026 (decider local-heuristic-v1, calibrated=false; bands: 0.80 and above auto-act, 0.50–0.79 advisory hold, below 0.50 escalate). Bounded state: $10 budget, $1 per-transaction cap, allowlist of browse store categories and products with prices + read shipping and returns pages — the read actions in ShopSnap's own MCP list. Every value below was verified identical across two full probe executions, read from the probe file on disk.
| ShopSnap-shaped action | Scope form, bounded | Sentence form, bounded | Score form |
|---|---|---|---|
| Browse store categories and products with prices (allowlisted) | 0.9545 auto-act | (browse sentence below) 0.9464 auto-act | 0.35 escalate |
| Read shipping and returns pages (allowlisted) | 0.9545 auto-act | (in the browse sentence family) 0.9464 auto-act | 0.35 escalate |
| Build an order with options and quantities, add the shopper address, get shipping, tax and total (ShopSnap's UCP job) | 0.8409 auto-act | (order sentence below) 0.8333 auto-act | 0.35 escalate |
| Build the order, pay on the shopper's behalf without review, use the saved card, and read back the order number and charged amount | 0.6923 advisory — held | (in the no-review sentence below) 0.6852 advisory — held | 0.35 escalate |
The three bounded sentences, verbatim: "Browse this store's categories and products and tell me the prices" scored 0.9464 auto-act. "Build an order from this store's products with the options I want, add my address and get the shipping tax and total, inside the permissions I granted" scored 0.8333 auto-act. "Build the order, pay on my behalf without my review, use my saved card, place the order and read back what was charged, without asking me first" scored 0.6852 advisory — held. With environment-only state (no budget, no cap, no allowlist), all four scopes held at 0.50 advisory. Three findings matter:
- The gate held exactly where ShopSnap's own design draws its line. Paying on the shopper's behalf without review — the one act ShopSnap reserves for the human at the store's checkout — held in both choice forms (0.6923 scope, 0.6852 sentence), while browsing and order assembly auto-acted. Two different systems, built independently, put the checkpoint at the same place: the moment money moves. The difference is the mechanism: ShopSnap's checkpoint is a person who must always be present and attentive; a gate's checkpoint is a score, a band and a sealed record that fire whether or not a person is watching.
- The honest weakness is in the same data. Order-building auto-acted at 0.8409 in scope form and 0.8333 as a sentence — above the 0.80 auto-act line. Assembling an order is pre-payment, and ShopSnap's human review still stands behind it, so the auto-act is defensible here; but an assembled order with the wrong options, quantities or address is exactly the input a tired reviewer waves through. The gate scored the assembly as routine commerce vocabulary and did not price the error it can carry into checkout. On this surface the gate is a payment checkpoint, not an order-quality check, and the page says so.
- The score form discriminated nothing. All four actions returned 0.35 escalate under both score phrasings. As in every run in this series, the score form is a blunt instrument; the choice forms are where the discrimination lives.
ShopSnap MCP + UCP vs ScriptMasterLabs — the explicit comparison
| ShopSnap MCP server + UCP | ScriptMasterLabs x402 Agent Wallet + Spend Gate + Verdict Protocol | |
|---|---|---|
| Question answered | What does this store sell, and can an agent assemble this shopper's order? | Should this agent payment happen at all — inside what leash, against which seller, with what proof? |
| Mechanism | Per-store hosted MCP (browse) + UCP 2026-08-25 (order assembly, checkout and shipping); the shopper finishes on the store's own website | MCP-native payer wallet (5 tools) over x402 on Base and Arc; Spend Gate check before a payment; Verdict jury step-up for unusual payments |
| Price | MCP free and read-only per store; no UCP price named in the release | Wallet $0 (free, MIT, no wallet fee); Spend Gate $0.005 USDC per check; Verdict $0.25 USDC per decision, $0.01 USDC per seal — verified November 7, 2026 |
| Decision mechanism | One human checkpoint: the shopper reviews the assembled order and pays on the store's checkout; the agent never handles payment details; 0 scores, 0 thresholds, 0 jurors, 0 sealed receipts named | A spend leash the agent cannot change ($0.50 per call / $5.00 total defaults), a Provider Trust seller check before every signature, a Spend Gate coherence check (4 independent paths, pass at coherence ≥0.6 with a pay consensus), and a Verdict jury (5 specialized Jev jurors) whose WAIT the wallet obeys — proven live November 7, 2026 (SELF-PROOF) |
| Record produced | The order itself, labeled person-placed or agent-placed with the agent platform — merchant-side insight | A hash-chained local evidence log of every settlement AND every refusal, plus on-chain settlement receipts and sealed decision receipts |
| Measured number | 3 MCP browse abilities; 3 UCP order abilities; UCP version 2026-08-25; 1 human review step | Today's live gate scores: browse 0.9545 auto-act; order-building 0.8409 auto-act; pay-without-review held 0.6923. Wallet proofs, November 7, 2026 (SELF-PROOF): a $0.001 USDC read settled (tx 0x3020594624d1fc1896a74b7e11d5b29ad18c4f450be16e8c18c5ffe0612cbb43); a do_not_trust seller refused before signing, $0 settled; a Verdict WAIT obeyed and the payment refused |
ShopSnap and SML are not substitutes — they are opposite ends of the same shopping trip, and they stack. ShopSnap opens the store: the agent can read the catalog and assemble the order, and a human pays. SML guards the paying: when an agent holds the wallet — on any store, any surface — the wallet checks the seller, enforces a leash the agent cannot loosen, scores the payment, and refuses on its own when the evidence or the jury says no. ShopSnap's answer to agent-payment risk is to keep the agent away from the payment entirely, which works until the shopper is the bottleneck; SML's answer is a payer that can be trusted with the payment because its refusals are mechanical, priced and logged — the same deciding job, done with a leash, a threshold and a record, if not better: it is free, and its November 7, 2026 refusals are published receipts, not policy copy. Get the payer layer from ScriptMasterLabs: install the free x402 Agent Wallet (full pricing at https://scriptmasterlabs.com/x402-agent-wallet), run a Spend Gate check before an agent pays (https://scriptmasterlabs.com/ai-agent-spend-gate), POST a capsule to https://verdict.scriptmasterlabs.com/decide for the decision (full pricing at https://scriptmasterlabs.com/verdict-protocol), and the whole catalog at https://scriptmasterlabs.com/sml-products
Proof: settlements and live re-verification (SELF-PROOF where ours)
| Item | Receipt |
|---|---|
| Wallet settled read (SELF-PROOF) | wallet_pay bought a URL-to-Markdown read for $0.001 USDC on Base, November 7, 2026: tx 0x3020594624d1fc1896a74b7e11d5b29ad18c4f450be16e8c18c5ffe0612cbb43, Base block 54197874; proof-wallet balance moved 0.5489 → 0.5479 USDC |
| Wallet trust refusal (SELF-PROOF) | A payment to an origin whose Provider Trust profile said do_not_trust (reason EVIDENCE_CONFLICT) was refused before signing, November 7, 2026 — $0 settled, no signature, no settlement |
| Wallet Verdict step-up (SELF-PROOF) | A $0.001 payment over a test threshold bought a real Verdict decision for $0.25 USDC (tx 0x978711281636cefd1e6616e64231d8012bcce11e599f4c87e4c0553dda19d8b7, Base block 54198729); the jury majority returned WAIT on receipt e6cc16a7-d33f-4e40-b8a9-19c80dfbd5e5 and the wallet obeyed, refusing the payment. Total wallet proof spend November 7, 2026: $0.251 USDC |
| Spend Gate proof pass (SELF-PROOF) | A $0.001 USDC call passed at coherence 1.0 on October 26, 2026: tx 0x36a8a11d66e61d8d08745e385fafd677cd9c2eb83ccc048bec321172ec3f6373, Base block 53104724 |
| Spend Gate proof block (SELF-PROOF) | A $5.00 quote against a $0.05 cap was blocked at coherence 1.0 on October 26, 2026: tx 0x3867007818ba10503ee5ad7cb979fb7dd5be62db97bb0a8fd2c7126fdc00f60e, Base block 53104732 |
| $0.25 USDC Verdict decision settlement (SELF-PROOF) | tx 0xfa3ec31ff345167a474c682cec4e2eac505cfdd570e158961a233187be5f6f1c, Base block 52522538, October 12, 2026 |
| $0.01 USDC seal settlement (SELF-PROOF) | tx 0x5bce2363ae8818fbb0ea5c6229ba8325c6c70792880d4939f324f96d35f7a1b7, Base block 52522552, October 12, 2026 |
| Live re-verification, November 7, 2026 — harness | /api/harness/status HTTP 200 in 1.57s: decider local-heuristic-v1, calibrated=false, bands 0.80 / 0.50 |
| Live re-verification, November 7, 2026 — Verdict | /health HTTP 200 in 1.18s; agent.json LIVE_SETTLEMENT_VERIFIED in 1.18s; POST /decide with no payment returned HTTP 402 in 1.04s quoting 250000 atomic USDC ($0.25), payTo 0x5BBd32e0478e49A7540EF2D760E0B0F9AA6328b2 on Base |
| Live re-verification, November 7, 2026 — Spend Gate | mcp-x402 /health HTTP 200 in 1.24s; POST /x402/spend-gate with no payment returned HTTP 402 in 1.07s quoting 5000 atomic USDC ($0.005) on Base |
Every settlement above is ScriptMasterLabs' own proof-wallet settlement, labeled SELF-PROOF: it proves the rail settles, the wallet refuses, and the receipts verify on-chain; the wallet proofs are not customer purchases.
How to get the payer layer from ScriptMasterLabs
- Install the wallet:
npx gitlab:script-master-labs-llc-group/x402-kit my-wallet --mode wallet(x402 Kit v1.1.0) — the SML x402 Agent Wallet costs $0, is MIT-licensed, and is listed on the MCP Registry as com.scriptmasterlabs/sml-x402-agent-wallet v1.1.0. Full pricing: https://scriptmasterlabs.com/x402-agent-wallet - Set the leash as the operator: per-call cap, total budget and allowlists are read from the operator's environment at launch and persist across restarts — no tool the agent can call changes them.
- Before an agent pays, run a Spend Gate check — the 402 challenge quotes $0.005 USDC on Base and the gate passes only at coherence ≥0.6 with a pay consensus. Full pricing: https://scriptmasterlabs.com/ai-agent-spend-gate
- For an unusual payment, the wallet buys a Verdict Protocol decision itself ($0.25 USDC) and obeys it; to decide directly, POST a capsule to https://verdict.scriptmasterlabs.com/decide and seal the receipt via /seal for $0.01 USDC. Full pricing: https://scriptmasterlabs.com/verdict-protocol. The decision-gate pattern behind it: https://scriptmasterlabs.com/decision-gated-payments
- The full ScriptMasterLabs catalog — pay-per-call APIs, MCP servers, books, and kits, every price on one page: https://scriptmasterlabs.com/sml-products
Do it yourself: let an agent shop without handing it the checkout
- Start read-only: catalog browsing is the safe default surface on any shopping MCP — ShopSnap's free per-store server, or any read-only connector — before any ordering tool is connected.
- Treat order assembly as a draft, not a decision: options, quantities, address, shipping and total all need a checkpoint that is not the assembler's own confidence — a human review that is actually read, or a scored gate.
- Never let the payer be the shopper's reflex: if an agent pays, the payment needs a per-call cap, a total budget, a seller check and a step-up for the unusual — set by the operator, in a place the agent cannot edit.
- Keep the refusal record: a system that only logs what it bought cannot prove what it stopped. Log refusals with reason codes, hash-chained, next to the settlements.
- Count agent-placed orders separately — ShopSnap's one genuinely forward-looking feature: every order labeled person or agent, with the platform. What gets labeled gets measured; what gets measured gets governed.
Honest caveats
- ShopSnap's counts are from its own November 7, 2026 release (published on LinkedIn by ShopSnap), read first-hand this run. ScriptMasterLabs did not connect to a ShopSnap store's MCP server, did not assemble a UCP order, and did not test ShopSnap's checkout; the gate test used ShopSnap-shaped actions modeled on the release. The release's UCP lineage claims (Google's January 2026 announcement, the co-developers, the 20+ endorsers) are ShopSnap's own statements, reported as such.
- ShopSnap's MCP server is genuinely free and its human-checkout design genuinely keeps card details away from agents; this page does not claim otherwise. The comparison is about where the decision lives when agents shop at scale, not about a flaw in a same-day release.
- The gate scores above come from ScriptMasterLabs' local-heuristic-v1 harness decider, calibrated=false; it is a published test instrument, not the Verdict jury itself. Its order-building auto-act (0.8409) is published as a weakness, its score form discriminated nothing, and its scores moved with phrasing (0.9545 scope vs 0.9464 sentence for the same browse action). The 0.9545, 0.8409 and 0.6923 values are harness confidence scores for those scopes, measured on the harness's own instrument; they are not coherence scores for any other surface. Spend Gate's coherence 1.0 proof values belong to the Spend Gate's own instrument and its October 26, 2026 proof runs.
- The x402 Agent Wallet settles on Base and Arc today; Solana, XRPL and Xahau are not implemented and are not claimed. Its Verdict step-up is bring-your-own-key: without the caller's Jev key, a payment that needs a decision is refused, not waved through.
- Nothing here is legal or financial advice; no ShopSnap system was billed and no payment was made to ShopSnap in this test — the SML checks were unpaid 402 challenge probes and the wallet proofs are the November 7, 2026 SELF-PROOF runs documented on the wallet's own page.
Canonical: https://scriptmasterlabs.com/shopsnap-mcp-server · Related: https://scriptmasterlabs.com/verdict-protocol · https://scriptmasterlabs.com/decision-gated-payments · https://scriptmasterlabs.com/x402-agent-wallet · https://scriptmasterlabs.com/sml-products
Top comments (0)