🎩 You're Invited:Meet the Socket team at Black Hat in Las Vegas, August 3-6.RSVP
Sign In

@marketbasketanalysis/mcp

Package Overview
Dependencies
Maintainers
1
Versions
1
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

@marketbasketanalysis/mcp

MCP server for MarketBasketAnalysis. 19 tools for AI shopping, merchant ops, B2B agents: recommendations, substitutes, cart bundle completion, rationale, cross-sell scoring, basket cohesion, return risk, reorder prediction, subscription kits, weekly plan,

latest
Source
npmnpm
Version
0.6.0
Version published
Maintainers
1
Created
Source

@marketbasketanalysis/mcp

npm version License MCP

An MCP server that gives any AI agent access to real co-purchase intelligence and merchant ops tooling from an ecommerce merchant's order history. 19 tools across discovery, bundle, insight, replenishment, merchant ops, and advanced mining. Works with Claude Desktop, Claude Code, Cursor, Windsurf, Cline, the OpenAI Agent SDK, and any other host that speaks the MCP stdio protocol. Works for merchants on Shopify, BigCommerce, WooCommerce, Magento, and OroCommerce. See "Platform coverage" below for which tools reach the self-hosted backends.

One npm package serves every marketplace. The server is platform agnostic, it is an HTTP client that calls a store's MBA backend over the public REST API. You re-point the whole server at any store with a single switch (MBA_API_BASE, see below). Most tools work on all five platforms; a handful depend on a backend route that not every platform ships yet. Per-tool marketplace coverage is the Tool catalog "Marketplace" column.

Why this exists

When a customer asks an AI shopping agent "what goes with the gym backpack?" the agent should give a real answer based on the merchant's actual order data, not a generic "you might also like" guess. When a merchant asks Claude "what should I work on this week?" the agent should pull from a ranked weekly plan, not invent tasks. This server makes both of those flows available to any MCP host in one line of config.

5-line install (Claude Desktop)

{
  "mcpServers": {
    "marketbasketanalysis": {
      "command": "npx",
      "args": ["-y", "@marketbasketanalysis/mcp"],
      "env": { "MBA_API_KEY": "mba_live_YOUR_KEY_HERE" }
    }
  }
}

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), restart Claude Desktop, the marketbasketanalysis server appears in the tools list with all 19 tools.

Point the server at your store (MBA_API_BASE)

The base URL is per-store configuration. By default the server talks to the shared hosted backend at https://app.marketbasketanalysis.com. If your data lives anywhere else, a BigCommerce store, a self-hosted backend, or a staging instance, set MBA_API_BASE so every tool reaches your own data plane:

{
  "mcpServers": {
    "marketbasketanalysis": {
      "command": "npx",
      "args": ["-y", "@marketbasketanalysis/mcp"],
      "env": {
        "MBA_API_KEY": "mba_live_YOUR_KEY_HERE",
        "MBA_API_BASE": "https://your-store-backend.example.com"
      }
    }
  }
}

MBA_API_BASE is the single switch that re-points the whole server; all 19 tools route through it. The value must be an https:// URL for non-local hosts (loopback, private, link-local, and metadata-service hosts are refused). For local development against a backend on localhost, set ALLOW_LOCAL_API_BASE=1 to allow an http://localhost base. Env-var changes take effect at server startup, so restart your MCP host after editing the value.

MBA_API_BASE per platform

The base URL is per-store configuration. Shopify, BigCommerce, and OroCommerce stores are served by the shared hosted backend, so they use the default. WooCommerce and Magento run the backend locally inside the store install, so point the server at the store's own domain:

PlatformMBA_API_BASE
Shopifyunset (hosted default https://app.marketbasketanalysis.com)
BigCommerceunset (hosted default)
OroCommerceunset (thin hosted client, same hosted backend)
WooCommercehttps://your-store.example.com (WordPress site URL). Its routes live under marketbasketanalysis/v1; the server maps paths automatically. See "Platform coverage" for which tools apply.
Magentohttps://your-magento.example.com (the /rest base). Its routes live under V1/marketbasketanalysis; the server maps paths automatically. See "Platform coverage" for which tools apply.

Platform coverage

The server writes canonical /api/v1/... paths and rewrites them per platform, because WooCommerce and Magento run the backend inside the store on their own REST conventions (marketbasketanalysis/v1 and V1/marketbasketanalysis respectively).

10 of the 19 tools reach WooCommerce and Magento: the six that derive from /recommendations (get_recommendations, get_bundle_for_cart, score_cross_sell, analyze_basket, propose_subscription_bundle, score_return_risk), plus find_substitutes, get_rationale, forecast_bundle, and predict_reorder.

The other 9 are the merchant-ops surface: get_opportunities, triage_opportunity, get_weekly_plan, execute_weekly_plan_action, get_drift_alerts, get_forecast_alerts, explain_opportunity, explain_drift, and mine_hui_itemsets. Those endpoints do not exist on the self-hosted backends. Calling one there returns a clear "not available on this platform" error naming the endpoint, with no network round trip, rather than an opaque 404.

For step-by-step install (config file location per OS, where to mint an API key, troubleshooting):

Authentication

The server reads MBA_API_KEY from the environment your MCP host passes in and sends it as a Bearer token on every request. To get a key:

  • Open the MarketBasketAnalysis admin (Shopify app drawer, or BigCommerce / WooCommerce / Magento / OroCommerce admin).
  • Click "API keys" in the left nav.
  • Click "Create key", name it, and copy the mba_live_ value (it is shown once).

Keys are per-shop, revocable, and rotated from the same screen. Only the SHA-256 hash is stored, so re-mint if a key leaks.

Auth model differs per marketplace, the MCP server abstracts it, but worth knowing:

  • Shopify, BigCommerce: Bearer mba_live_... straight through. This is the common path.
  • WooCommerce: Bearer against a Woo-minted key, which must carry the customer_data scope for predict_reorder.
  • Magento: tools reach the store over the Magento REST surface (/V1/marketbasketanalysis/* and /V1/mba/*); some routes are admin-token / ACL scoped on the store side.
  • OroCommerce: the store sits behind the platform OAuth2 firewall for /api/ routes; the hosted backend the thin client proxies to is what the MCP server actually calls, so the mba_live_ key still applies.

Tool catalog

19 tools, organized into the four Basket AI agent roles plus two operational groups. The Marketplace column states which backends ship the route the tool calls, which is not the same as which backends this server can currently REACH: see "Platform coverage" above. "All five" means Shopify, BigCommerce, WooCommerce, Magento, OroCommerce.

Discovery

ToolDescriptionRequired paramsMarketplace
get_recommendationsComplementary products for a single product.product_idAll five
find_substitutesReplacement options when a product is unavailable.product_idAll five
get_rationaleOne-sentence "why" for a recommendation pair.product_id, related_product_idAll five

Bundle

These derive everything from /recommendations (the server composes the bundle/scoring logic client-side), so they need no extra backend route and work everywhere.

ToolDescriptionRequired paramsMarketplace
get_bundle_for_cartMissing kit components for a multi-item cart.product_idsAll five
propose_subscription_bundleRecurring subscription kit proposal.seed_product_idsAll five

Insight

Also /recommendations-derived, so universal.

ToolDescriptionRequired paramsMarketplace
score_cross_sellStrength verdict for a (a, b) pair.product_a, product_bAll five
score_return_riskBundle return-risk score.product_idsAll five
analyze_basketCohesion score for a proposed bundle.product_idsAll five

Replenishment + forecasting

ToolDescriptionRequired paramsMarketplace
predict_reorderB2B reorder cadence per customer / SKU.customer_idShopify, BigCommerce, WooCommerce, Magento. Hidden when MBA_PLATFORM=orocommerce.
forecast_bundleWeekly Holt-Winters forecast + buy quantity.bundle_idShopify, BigCommerce, Magento (/forecast/bundle-inventory). Not on OroCommerce.

Merchant ops

These call Bearer /api/v1 routes that ship on BigCommerce today. Shopify serves opportunities, drift, and the weekly plan through its embedded admin views rather than an /api/v1 route, so these tools resolve against a BigCommerce backend. The one exception is /explain-opportunity, which now ships on BigCommerce and Shopify; /explain-drift remains BigCommerce only. The tools surface a clean upstream 404 on platforms that lack the route.

ToolDescriptionRequired paramsMarketplace
get_weekly_planRanked weekly action list.(none)BigCommerce
execute_weekly_plan_actionDispatch a specific action (confirm-gated).action_id, confirmBigCommerce
get_opportunitiesMined opportunities, ranked.(none)BigCommerce
explain_opportunityStats (support / confidence / lift / sample count) plus a templated "why this is a good cross-sell" narrative for one opportunity.opportunity_idBigCommerce, Shopify
triage_opportunityActivate / pause / archive (confirm-gated).opportunity_id, action, confirmBigCommerce (POST /opportunities/{id}/action); admin grid on other platforms.
get_drift_alertsRules whose confidence has drifted.(none)BigCommerce
explain_driftStats plus a templated "why this pair drifted" narrative for one drift alert (degrades gracefully for a disappeared pair).alert_idBigCommerce
get_forecast_alertsBundles at risk of stockout / demand drop.(none)BigCommerce

Advanced mining

ToolDescriptionRequired paramsMarketplace
mine_hui_itemsetsHigh-utility itemset mining (Plus / Enterprise).ordersShopify, BigCommerce, WooCommerce, OroCommerce. Plus / Enterprise tier.

Example prompts per tool

Paste any of these into a Claude Desktop / Claude Code / Cursor chat after wiring up the server:

  • get_recommendations: "Use marketbasketanalysis to find what customers also buy with the gym backpack (product 8472918765)."
  • find_substitutes: "The DSLR body is out of stock. What is a good substitute?"
  • get_rationale: "Why is the water bottle recommended with the gym backpack?"
  • get_bundle_for_cart: "I have a camera body, 32GB SD card, and a tripod in my cart. What is likely missing to make this a complete kit?"
  • propose_subscription_bundle: "Build a monthly subscription kit for customer 9876."
  • score_cross_sell: "Is a cleaning kit a good cross-sell for the DSLR camera body?"
  • score_return_risk: *"What is the return risk of the camera + lens
    • tripod + bag bundle?"*
  • analyze_basket: "I am thinking of bundling camera + lens + SD card + bag. Based on actual customer data, is that a strong bundle?"
  • predict_reorder: "What is Acme Corp (customer 7654321) due to reorder this week?"
  • forecast_bundle: "Forecast bundle b-camera-kit for the next 12 weeks and recommend a buy quantity."
  • get_weekly_plan: "What is on my weekly plan?"
  • execute_weekly_plan_action: "Run action a-42 from my weekly plan, confirmed."
  • get_opportunities: "Show me my top three proposed opportunities."
  • explain_opportunity: "Why is opportunity opp-17 a good cross-sell?"
  • triage_opportunity: "Activate opportunity opp-17, confirmed."
  • get_drift_alerts: "Are any of my rules drifting?"
  • explain_drift: "Why did the pair in drift alert alert-7 drift?"
  • get_forecast_alerts: "Which bundles are at risk of stockout?"
  • mine_hui_itemsets: "Mine top-20 high-utility itemsets from this 90-day order payload." (Plus / Enterprise tier)

Per-tool narrative docs live in the cookbook.

Env vars

VariableRequiredDefaultNotes
MBA_API_KEYyes--The mba_live_... key from your admin
MBA_API_BASEnohttps://app.marketbasketanalysis.comPer-store base URL. Set this for BigCommerce, self-hosted, or staging backends so the server points at your data plane. Must be https:// for non-local hosts.
MBA_PLATFORMno(any)Set to shopify or bigcommerce to expose platform-gated tools (e.g. predict_reorder)
MBA_SENTRY_DSNno--Opt-in error telemetry (merchant-controlled)
MBA_DEBUG_ERRORSno--Set to 1 to print upstream error bodies to stderr
ALLOW_LOCAL_API_BASEno--Set to 1 to permit localhost in MBA_API_BASE during dev

Development

git clone https://github.com/48x-ai/marketbasketanalysis
cd marketbasketanalysis
pnpm install
pnpm --filter ./packages/mcp typecheck
pnpm --filter ./packages/mcp test
pnpm --filter ./packages/mcp dev    # tsx-based local run
pnpm --filter ./packages/mcp build  # emit ./dist

Adding a new tool

Each tool is a self-contained module under src/tools/. To add one:

  • Create src/tools/myNewTool.ts exporting definition and handler. Mirror the structure of src/tools/getRecommendations.ts for a simple GET, or src/tools/triageOpportunity.ts for a POST with a confirm gate.
  • Register it in src/tools/index.ts by importing and adding the module to the allModules array.
  • Add tests in src/tools/myNewTool.test.ts, covering: missing-key reply, happy path, and at least one upstream-error path. Mirror src/tools/findSubstitutes.test.ts.
  • Document it in the table above and in dist/mcp/smithery.yaml.

Distribution artifacts

The dist/mcp/ directory at the monorepo root holds the install samples (Claude Desktop, Cursor, Windsurf), the Smithery YAML, and the Anthropic marketplace submission content. See dist/mcp/README.md for the full layout.

Publishing

Tag-based publish via the monorepo-root .github/workflows/publish-mcp.yml workflow. Bump the version in packages/mcp/package.json (keep server.json and src/index.ts in sync), merge to main, then tag mcp-v$VERSION and push the tag. The workflow runs typecheck, test, build, a tag/version match check, then npm publish --access public --provenance using the NPM_TOKEN repo secret. A workflow_dispatch manual trigger is available for rescue runs.

The full operator checklist, including the one-time NPM_TOKEN setup and a no-CI manual publish fallback, lives in docs/RELEASE.md.

Troubleshooting

SymptomCause / fix
Server does not appear in tool drawerJSON typo or npx not on PATH. Check the host's MCP log.
"Error: MBA_API_KEY environment variable not set"env block missing or value empty.
"MBA API 401"Key revoked or wrong; mint a fresh one.
"MBA API unreachable"Network reach failed; check https://status.marketbasketanalysis.com.
Tool times out on first callFirst npx -y cold-start downloads the package; install globally for repeat speed.
"MBA API returned malformed response"Upstream backend drift; set MBA_DEBUG_ERRORS=1 to see the body in stderr.
predict_reorder missingThe tool is platform-gated: it registers when MBA_PLATFORM is unset, shopify, or bigcommerce. The reorder-prediction route also exists on WooCommerce and Magento but the tool gate does not expose it there yet.

For deeper diagnostics see each per-IDE setup doc under dist/mcp/.

License

UNLICENSED, proprietary.

Keywords

mcp

FAQs

Package last updated on 05 Aug 2026

Did you know?

Socket

Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.

Install

Related posts