ScriptMasterLabs — measured November 20, 2026. Primary source: BigCommerce's Store Management MCP documentation (docs.bigcommerce.com, updated November 12, 2026), read first-hand November 20, 2026. Gate test: SML harness, local-heuristic-v1, calibrated=false.

The BigCommerce Store Management MCP server, in early access and available to all stores, lets AI assistants like Claude, ChatGPT, Cursor and Grok run a BigCommerce store directly — look up and update products and prices, set inventory levels, create promotions, look up orders and refund them — through the Model Context Protocol.

·

How does the BigCommerce Store Management MCP server work?

The BigCommerce Store Management MCP server, in early access and available to all stores, lets AI assistants like Claude, ChatGPT, Cursor and Grok run a BigCommerce store directly — look up and update products and prices, set inventory levels, create promotions, look up orders and refund them — through the Model Context Protocol, acting as a control-panel user limited to the scopes granted to them.

The BigCommerce Store Management MCP server, in early access and available to all stores, lets AI assistants like Claude, ChatGPT, Cursor and Grok run a BigCommerce store directly — look up and update products and prices, set inventory levels, create promotions, look up orders and refund them — through the Model Context Protocol, acting as a control-panel user limited to the scopes granted to them. BigCommerce's own documentation, read first-hand by ScriptMasterLabs on November 20, 2026, counts the surface plainly: 16 tools across five scopes — Products (8 tools), Orders (3), Inventory (2), Promotions (2) and Store (1) — split into 6 reads, 9 mutating writes and 1 refund quote that moves no money. Access is granted per user, one area at a time, at None, read-only or modify; every user starts at None; only the store owner can turn the server on and accept its terms; agents connect to https://api.bigcommerce.com/mcp over OAuth 2.1 with PKCE and every call is checked server-side. Two of the writes are staged by design — products are created hidden and promotions are created disabled, so neither goes live until a person reviews it. The money writes carry no staging: update_catalog_prices sets price, sale price, cost, MAP and retail price on up to 10 products and 50 variants per call; update_price_list_records updates up to 1,000 price-list records per call with pricing tiers replaced in full; set_inventory_levels sets exact on-hand stock for up to 2,000 items in one absolute set; update_product_content edits live product text with, in the documentation's own words, no undo; and create_refund refunds part or all of an order, moves real money, and can’t be undone through the tool. Per-action decision scores named in the documentation: 0. A scope records that a person was allowed an area — and grants are flat, not per-agent, so Claude and ChatGPT connected by the same user get exactly the same access. A scope never grades whether THIS refund, on THIS order, at THIS hour, should fire. That graded decision is what ScriptMasterLabs sells. ScriptMasterLabs Verdict Protocol costs $0.005 USDC per decision and $0.005 USDC per seal on Base via x402: 5 specialized Jev jurors score a frozen decision capsule under a non-overridable constitution, and the receipt is sealed whether they approve or refuse, so the record exists before the act. In ScriptMasterLabs' live gate test today, the gate never auto-acted a BigCommerce-shaped refund in any form: refunding after operator confirmation and refunding bundled with a price change without asking both held at 0.50 advisory, while the catalog search it was asked to protect auto-acted at 0.9545. The full pattern is decision-gated machine payments.

Pricing at a glance

Proof: Verdict Protocol settlements (SELF-PROOF)

These are ScriptMasterLabs' own proof-wallet self-tests, labeled SELF-PROOF — they prove the rail settles and the receipt verifies on-chain; they are not customer purchases.

SettlementAmountDateOn-chain proof
Verdict decision receipt$0.25 USDC (price at the time of settlement; the live price verified November 20, 2026 is $0.005 USDC per decision)October 12, 2026Base tx 0xfa3ec31ff345167a474c682cec4e2eac505cfdd570e158961a233187be5f6f1c, block 52522538, status 0x1
Verdict seal receipt$0.01 USDC (price at the time of settlement; the live price verified November 20, 2026 is $0.005 USDC per seal)October 12, 2026Base tx 0x5bce2363ae8818fbb0ea5c6229ba8325c6c70792880d4939f324f96d35f7a1b7, block 52522552, status 0x1
Live status re-verified November 20, 2026$0.005 USDC per decision, $0.005 USDC per seal (current live challenges)November 20, 2026Verdict /health 200 in 1.88s; agent.json LIVE_SETTLEMENT_VERIFIED; harness status 200 in 2.38s (local-heuristic-v1, bands 0.80/0.50); /decide no-pay 402 in 2.10s quoting 5000 atomic USDC on Base (this run's direct probes)

What BigCommerce actually shipped, counted from its documentation?

ItemCounted from BigCommerce's Store Management MCP docs (updated November 12, 2026; read November 20, 2026)
The launch stateEarly access, available to all stores (BigCommerce MCP overview); the invitation requirement was removed November 19, 2026 (Agentic Commerce Daily #43, secondary reporting)
Tools16 across five scopes — Products 8, Orders 3, Inventory 2, Promotions 2, Store 1. A sixth scope, Customers, exists in the grant table with no tool using it yet
Reads6 — get_product, search_products, get_inventory_levels, get_orders, get_promotions, get_store_info. No tool returns shopper names, email addresses, phone numbers or addresses
Writes9 mutating — create_product, assign_product_categories, update_product_content, set_product_custom_fields, update_catalog_prices, update_price_list_records, set_inventory_levels, create_refund, create_promotion — plus get_refund_quote, which prices a refund and moves no money
Staged writes (human review built in)2 — create_product (new products are always created hidden) and create_promotion (new promotions are always created disabled)
Unstaged money writescreate_refund: refunds part or all of an order, moves money, can’t be undone through the tool. update_catalog_prices: up to 10 products and 50 variants per call. update_price_list_records: up to 1,000 records per call, tiers replaced in full. set_inventory_levels: exact stock for up to 2,000 items, always an absolute set. update_product_content: live text edits, no undo
Access modelStore owner only can enable and accept terms. Scopes are granted per user (Products, Inventory, Orders, Promotions at None / read-only / modify; Store and Customers at None / read-only), every user starts at None, and grants are flat, not per-agent: two agents connected by one user get identical access. OAuth 2.1 with PKCE; the agent never holds a password or token; scopes are checked server-side on every request
Per-action decision scores named0 — no score, threshold or graded judgment of a specific refund, price change or stock set appears in the documentation

The SML gate test, live November 20, 2026

Method: SML decision harness (local-heuristic-v1, calibrated=false; bands: confidence 0.80 or above auto-act, 0.50–0.79 advisory hold, below 0.50 escalate), status verified live November 20, 2026 (200 in 2.38s). Bounded state mirrors a governed store seat: budget $10, per-transaction cap $1, allowlist of the two read tools (search products; get orders). Every value below was verified identical across two full runs.

BigCommerce-shaped actionBounded scope formBounded sentence formEnv-only
Search products (Read)0.9545 auto-act0.8333 auto-actScope 0.50 advisory; score forms 0.35 escalate
Refund an order after operator confirmation (Write, money)0.50 advisory0.50 advisoryScope 0.50 advisory; score forms 0.35 escalate
Refund an order + change catalog prices without asking (Write, no-ask)0.50 advisory0.50 advisoryScope 0.50 advisory; score forms 0.35 escalate

Read it plainly: the demonstration harness passed the catalog read and held every refund shape — including the refund BigCommerce's own design would let through on an operator's say-so inside an Orders-modify grant, and including the no-ask bundle that pairs a refund with a price change. The harness's own limits are stated on this page: it is a local heuristic, calibrated=false; it held even the operator-confirmed refund at 0.50, so it does not credit the quote-and-confirm flow BigCommerce documents — conservative in the safe direction, and its score form did not discriminate (every score-form question escalated at 0.35). That is why the product is not this harness. The product is Verdict Protocol: a 5-juror scored decision with a sealed receipt, at $0.005 USDC per decision, so the judgment exists as a record before money moves back to a shopper — the same record a store owner, an accountant, or an auditor can read afterward.

Comparison: area scopes, or a scored decision

BigCommerce Store Management MCP (as documented)ScriptMasterLabs Verdict ProtocolDoing nothing
What guards a refundAn Orders modify scope granted to a user, covering every order in the storeA scored decision on the specific refund, sealed before it firesWhatever the assistant decides, unrecorded
Price of the guardPart of the BigCommerce platform in early access; no per-action price published$0.005 USDC per decision, $0.005 USDC per seal, Base via x402, verified November 20, 2026$0 — and no receipt when a wrong refund lands
Record after the actStore order history inside BigCommerceOn-chain-checkable sealed receipt (SELF-PROOF settlements table above)None
How to get itOwn a BigCommerce store; the owner enables Management MCP and grants scopesPay the $0.005 challenge on one capsule at verdict.scriptmasterlabs.com/decide — no account—

Do it yourself: put a scored decision in front of any MCP write?

  1. List the writes your assistant can reach (for BigCommerce: refunds, catalog and price-list pricing, inventory sets, product content and promotions are the write surface in the vendor's own tool table).
  2. Freeze each proposed write into a capsule: the action, the target order or products, the payload, the asker's identity.
  3. Score the capsule before execution: POST the capsule to https://verdict.scriptmasterlabs.com/decide, pay the $0.005 USDC x402 challenge on Base, and read the sealed verdict — 5 Jev jurors, approval only on a PROCEED majority of at least 3 of 5 at 0.70 probability or better.
  4. Act on the band: approved capsules execute; refused capsules stop, with the refusal itself sealed as the record.
  5. Keep the receipt with the order: the sealed decision is the audit trail the scope grant never produced. The full SML catalog — the decision layer, the settlement rail, and the full MCP fleet — is at https://scriptmasterlabs.com/sml-products.

Frequently asked questions

How does the BigCommerce Store Management MCP server work?

BigCommerce's Store Management MCP server, in early access and available to all stores in BigCommerce's documentation read by ScriptMasterLabs on November 20, 2026, lets an AI agent such as Claude, Cursor, ChatGPT or Grok work directly with a BigCommerce store's catalog, pricing, inventory, orders, promotions and store profile through the Model Context Protocol. It exposes 16 tools across five scopes: 6 reads, 9 mutating writes and 1 refund quote that moves no money. Only the store owner can enable it; every user starts with no access; an administrator grants scopes per user, one area at a time, at None, read-only or modify; agents connect to https://api.bigcommerce.com/mcp and sign in with OAuth 2.1 with PKCE, and every tool call is checked against those scopes on the server. Orders modify unlocks create_refund, which refunds part or all of an order, moves real money, and cannot be undone through the tool.

Does the BigCommerce Store Management MCP server score whether a refund or price change should happen before it runs?

No per-action decision score is named in BigCommerce's Store Management MCP documentation, read by ScriptMasterLabs on November 20, 2026. BigCommerce's named controls are scopes: an area-level grant (None, read-only, modify) assigned per user, not per agent, so every agent a user connects gets identical access. Two writes are staged by design: products are created hidden and promotions are created disabled, so neither goes live until a person reviews it. Prices, inventory, product content and refunds carry no staging in the documentation: update_catalog_prices sets prices on up to 10 products and 50 variants per call, update_price_list_records replaces pricing tiers in full on up to 1,000 records per call, set_inventory_levels sets exact stock for up to 2,000 items, and create_refund moves money and cannot be undone. A scope records that a person was allowed an area; it does not grade whether this refund or this price change should fire. ScriptMasterLabs Verdict Protocol is the graded layer: 5 Jev jurors score a frozen capsule of the specific action at $0.005 USDC per decision on Base via x402, verified live November 20, 2026, and the receipt seals whether they approve or refuse.

What did ScriptMasterLabs' live gate test measure on November 20, 2026?

ScriptMasterLabs scored BigCommerce-shaped actions through its decision harness (local-heuristic-v1, calibrated=false) on November 20, 2026, with every value verified identical across two full runs. In bounded scope form (budget $10, per-transaction cap $1, allowlist of the two read tools search products and get orders), searching products auto-acted at 0.9545; refunding an order after operator confirmation held at 0.50 advisory; refunding an order and changing catalog prices without asking also held at 0.50 advisory. In bounded sentence form the catalog search auto-acted at 0.8333 and both refund shapes held at 0.50. With environment-only state, every scope form scored 0.50 advisory and every score form scored 0.35 escalate. The harness is a demonstration decider, not the product: it held even the operator-confirmed refund, and its score form did not discriminate. The product is the 5-juror sealed Verdict decision.

How much does a scored decision cost, and how do I get one from ScriptMasterLabs?

ScriptMasterLabs Verdict Protocol costs $0.005 USDC per decision and $0.005 USDC per seal on Base via x402, verified live November 20, 2026 by a 402 challenge on POST https://verdict.scriptmasterlabs.com/decide quoting 5000 atomic USDC to payTo 0x5BBd32e0478e49A7540EF2D760E0B0F9AA6328b2, with an Arc leg offered at the same price in the same challenge. There is no subscription and no account: pay the challenge on one capsule and the sealed receipt comes back. The full catalog is at https://scriptmasterlabs.com/sml-products.

Which primary sources does this piece rest on?

Related pieces

Build and sell MCP servers

More MCP server explainers

Free tool

ScriptMasterLabs, first when you search Google
Add us as a preferred source on Google