How does the Virid marketAgility MCP connector work?

Virid's marketAgility MCP Connector, launched October 28, 2026, connects a retailer's marketAgility store to AI assistants such as Claude and ChatGPT, so the store team can search 8 named record types — products, orders, customers, shopping carts, categories, inventory, promotion codes and tax nexus records — and create or update most of them in plain language, while orders, carts and per-site inventory stay read-only and every proposed change is shown for review before it is made. ScriptMasterLabs' live decision gate scored the surface the same day and reproduced Virid's own line in both test forms: the bounded read auto-acted (0.9545 in scope form, 0.8333 as the vendor's example sentence), and every write held for review at 0.50 — including the vendor's own worked example, "Create a new promotion code for 15% off, expiring at the end of the month." A 15% promotion code is a margin decision, not bookkeeping; in this test the scored gate and the vendor's review step agreed on exactly where it belongs.

Correction note (October 28, 2026): this page was first published with gate values taken from an unverified reading of the probe run. Every value below was re-read from the probe file on disk (probe-virid-1028.out, two consecutive runs, identical) and the page corrected the same morning. The first version's sentence-form promotion score (0.9321 auto-act) was wrong; the verified sentence-form score is 0.50 advisory, a hold.

What Virid shipped on October 28, 2026

The launch facts, counted from the October 28, 2026 PR Newswire release (primary, read October 28, 2026):

Item	Count / value
Searchable record types named	8 — products, orders, customers, shopping carts, categories, inventory, promotion codes, tax nexus records (release adds "and more")
Write surface	Create or update most of the named records
Read-only classes	3 — orders, shopping carts, per-site inventory
Worked examples in the release	2 — find products with SKUs starting with WDG not updated in the last 90 days; create a new promotion code for 15% off expiring at the end of the month
Named control	1 universal step — users review any proposed change before it is made
Authentication	OAuth 2.1; users authenticate directly with marketAgility, not with the AI platform; no API tokens or technical configuration required at setup
Data handling named	Connector does not store credentials, does not retain store data, all traffic encrypted in transit
Availability	All Virid clients with an active marketAgility store and a marketAgility account with API access; works with any AI platform supporting custom MCP connectors, including Claude and ChatGPT (availability varies by platform and plan)
Vendor	Virid, Unified Commerce provider, operating since 1999; CEO Steve Deller

The money anatomy matters here. A promotion code is a price cut the store has agreed to honour on every qualifying order until it expires — creating one commits margin. Updating catalog and pricing records changes what customers are charged. Tax nexus records are compliance data. Virid's design recognises the risk in one specific way: the human reviews any proposed change before it is made. That is a visibility step — the reviewer sees the change — not a judgment step: nothing scores whether the instruction behind the change was sound, bounded, or even the instruction the merchant meant to give. The gate test below measures whether a scored layer draws the same line Virid drew by design.

The live gate test — October 28, 2026, ~09:19 ET

ScriptMasterLabs ran its public harness gate against Virid-shaped actions. The decider is local-heuristic-v1, calibrated=false, bands ≥0.80 auto-act, 0.50–0.79 advisory hold, <0.50 escalate (status verified live the same minute). The bounded state was a $10 budget, a $1 per-transaction cap, and an allowlist of 2 actions: search products, orders and customers; find stale products by SKU. Every value below was verified identical in two consecutive runs, read from the probe file on disk.

Form	Search products, orders, customers (read)	Create a new promotion code (write)	Update product catalog and pricing records (write)
Choice, scope form, bounded state	0.9545 auto-act	0.50 advisory	0.50 advisory
Choice, scope form, environment only	0.50 advisory	0.50 advisory	0.50 advisory
Choice, full sentence, bounded state	0.8333 auto-act	0.50 advisory	0.50 advisory
Score form, environment only (both phrasings)	0.35 escalate	0.35 escalate	0.35 escalate

Read it plainly:

• In scope form with bounded state the gate drew Virid's line exactly: the read auto-acted at 0.9545, and both writes — promotion-code creation and the catalog/pricing update — held at 0.50 advisory. No write auto-executed.
• The sentence form held the same line, which this series does not take for granted: the stale-product read — the release's first worked example, verbatim — auto-acted at 0.8333; the promotion creation — the release's second worked example, verbatim — held at 0.50 advisory; and the destructive bundle phrased with "without asking first" (update prices and delete the old promotion codes) also held at 0.50. Sentence phrasing moved the read's score (0.9545 scope vs 0.8333 sentence) but flipped no decision.
• The bounded context did the read's work: with environment-only state the same read collapsed to 0.50 advisory, level with the writes — the bounded-context A/B replicated again (allowlist + budget + cap in the state is what earns the auto-act; without them the gate holds everything, reads included).
• The score form did not discriminate at all: every action, including the harmless read, escalated at 0.35. The gate's discrimination lives in the choice form only — published as a limit, not hidden.

The synthesis: Virid shipped the review step the whole October MCP series has been converging on — every proposed change is shown before it is made — and in this test a scored gate independently reproduced that step's line in both forms, which is the strongest vendor/gate alignment the series has recorded alongside the Airbyte scope form. The remaining gap is not the line, it is the reason: Virid's reviewer sees the change but gets no score, no band and no sealed record of why this instruction was allowed to propose it. A scored decision layer supplies exactly that record in front of the review step, not instead of it.

Where the decision layer fits

Creating a promotion code is decision-gated machine payment logic in retail dress: a confidence judgment ("this discount, at this depth, for this long, is what the merchant intends") is what stands between a sentence and committed margin. ScriptMasterLabs' Verdict Protocol is the scored-decision layer that sits before actions like these fire: a 5-juror decision for $0.25 USDC on Base, with the instruction, confidence, band and outcome sealed into a receipt — so "create a 15% code" carries the review record that justified it, not just the review screen that displayed it.

Same-day live re-verification (October 28, 2026): Verdict /health returned 200 in 1.70s, its agent.json read LIVE_SETTLEMENT_VERIFIED in 1.78s, and a no-payment /decide probe returned HTTP 402 in 1.23s quoting 250,000 atomic units ($0.25 USDC, Base).

Do it yourself

1. Start with the read-only classes as your safety map: orders, carts and per-site inventory can never be changed through the connector — use them freely for questions ("find products with SKUs starting with WDG that haven't been updated in the last 90 days") while you learn what the assistant proposes on the writable surface.
2. Treat promotion codes as money, because they are: before granting create/update access, decide your discount ceiling and expiry discipline in writing, and require the review step to show the code's depth, scope and expiry every time — a 15% code expiring at month-end is a different decision from the same code with no expiry.
3. Score the sentence, not just the scope: before relying on the review step, run your real workflow sentences — above all the vendor's own "Create a new promotion code for 15% off, expiring at the end of the month" — through a decision gate (≥0.80 auto-act, 0.50–0.79 confirm, <0.50 escalate) in both scope form and the full sentence your team would actually type, with your bounded state (budget, per-transaction cap, allowlist) in place. Here both forms held the creation at 0.50 and auto-acted only the reads — verify your own surface the same way, because in this series other vendors' sentence forms have flipped holds into auto-acts.
4. Keep tax nexus and pricing updates in a separate, slower lane: they are named writable record types in this connector, they carry compliance and revenue consequences a general review click is not designed to catch, and they are exactly the updates to route through a second approver until your gate receipts show a clean history.

Honest limits

• The gate decider is a local heuristic (local-heuristic-v1, calibrated=false); TypeSafe/Jev is not wired into this harness. The scores measure this gate's behaviour, not Virid's platform, and no Virid MCP call was made — the capability counts are the launch release's, read October 28, 2026.
• The release does not enumerate individual tools or a total tool count, so none is claimed here; the counted surface is the 8 named searchable record types and the 3 named read-only classes. The release's searchable list ends with "and more," so 8 is a floor, not a total.
• The release names one universal control — review of any proposed change before it is made — and does not state whether any record type is exempt from that review, or whether deletion is exposed at all; the gate test's delete phrasing was SML's probe, not a vendor-exposed tool claim.
• Whether any retailer has yet had an assistant create a promotion code in production couldn't verify.

Canonical: https://scriptmasterlabs.com/virid-marketagility-mcp-connector Related: https://scriptmasterlabs.com/verdict-protocol · https://scriptmasterlabs.com/decision-gated-payments · https://scriptmasterlabs.com/ai-agent-spend-gate · https://scriptmasterlabs.com/the-fourth-layer