@dcluttr/dclare-mcp 0.1.6 → 0.1.8

package/context/datasets.json

@@ -13057,7 +13057,7 @@
     },
     {
       "name": "blinkit_sku_insights",
-      "description": "SKU-level performance insights on Blinkit, including sales, inventory, availability, pricing, visibility, share of voice, and market share metrics.",
+      "description": "SKU-level performance insights on Blinkit, including sales, inventory, availability, pricing, visibility, share of voice, and market share metrics. Use Average metric if available instead of weighted metrics",
       "grain": "One row per date, brand, and product.",
       "owner": "BLINKIT Analytics",
       "timeDimension": "created_at",
@@ -13420,17 +13420,17 @@
         },
         {
           "name": "avg_ds_listing_pct",
-          "description": "Average dark store listing percentage.",
+          "description": "Use this when asked for listing information. Average dark store listing percentage.",
           "aggregation": "sum"
         },
         {
           "name": "avg_osa_ls_pct",
-          "description": "Average on-shelf availability in listed stores.",
+          "description": "Use this when asked for on-shelf availability information in listed stores. Average on-shelf availability in listed stores.",
           "aggregation": "sum"
         },
         {
           "name": "avg_osa_pct",
-          "description": "Average on-shelf availability percentage across all dark stores.",
+          "description": "Use this when asked for on-shelf availability information across all stores. Average on-shelf availability percentage across all dark stores.",
           "aggregation": "sum"
         },
         {
@@ -13440,22 +13440,22 @@
         },
         {
           "name": "wt_osa_ls_pct",
-          "description": "Weighted OSA in listed stores weighted by sales.",
+          "description": "Weighted OSA in listed stores weighted by sales. don't use this metric",
           "aggregation": "sum"
         },
         {
           "name": "wt_osa_pct",
-          "description": "Weighted on-shelf availability percentage across all dark stores.",
+          "description": "Weighted on-shelf availability percentage across all dark stores. don't use this metric",
           "aggregation": "sum"
         },
         {
           "name": "wt_discount_pct",
-          "description": "Weighted discount percentage across all relevant products weighted by sales.",
+          "description": "Weighted discount percentage across all relevant products weighted by sales don't use this metric.",
           "aggregation": "sum"
         },
         {
           "name": "wt_ppu",
-          "description": "Weighted price per unit (x100) — sales-weighted average PPU across products.",
+          "description": "Weighted price per unit (x100) — sales-weighted average PPU across products. don't use this metric",
           "aggregation": "sum"
         },
         {

package/dist/index.js

@@ -133,9 +133,12 @@ server.registerTool("get_dataset_context", {
 server.registerTool("run_semantic_query", {
     title: "Run semantic query",
     description: "Runs a semantic query against the data platform with guardrails. " +
-        "IMPORTANT: Brand-specific datasets contain both the user's brand AND competitor data. " +
+        "IMPORTANT RULES: " +
+        "1. TIME DIMENSIONS: For any query involving a date range or daily/weekly/monthly trend, you MUST use the timeDimensions parameter with the dataset's time column (usually 'created_at'). Use dateRange for filtering (e.g. 'this month', 'last 7 days', ['2026-02-01','2026-02-28']) and granularity ('day','week','month') for time-series breakdowns. " +
+        "2. BRAND FILTER: Brand-specific datasets contain both the user's brand AND competitor data. " +
         "When the user asks about their own brand's performance, include the 'curr_brand' segment to filter to their brand only. " +
-        "When the user asks for competitor analysis or market-wide comparison, do NOT include 'curr_brand' so all brands are returned.",
+        "When the user asks for competitor analysis or market-wide comparison, do NOT include 'curr_brand' so all brands are returned. " +
+        "3. MEMBER NAMES: Use plain column names (e.g. 'created_at', 'offtake_mrp') without the dataset prefix.",
     inputSchema: {
         tenantToken: z.string().optional(),
         query: z.object({
@@ -168,7 +171,8 @@ server.registerTool("ask_data_question", {
         "IMPORTANT: Always call search_datasets first to resolve the exact dataset name and pass it as datasetHint. Do NOT guess dataset names. " +
         "Brand-specific datasets contain both the user's brand AND competitor data. " +
         "When the question is about the user's own brand, the plan should include the 'curr_brand' segment. " +
-        "When the question involves competitor analysis or market-wide data, do NOT include 'curr_brand'.",
+        "When the question involves competitor analysis or market-wide data, do NOT include 'curr_brand'. " +
+        "For any question involving dates, trends, or time periods, the plan MUST include timeDimensions with the correct dateRange and granularity.",
     inputSchema: {
         tenantToken: z.string().optional(),
         question: z.string().min(5),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dcluttr/dclare-mcp",
-  "version": "0.1.6",
+  "version": "0.1.8",
   "type": "module",
   "description": "MCP server for secure talk-to-data on Cube + ClickHouse",
   "bin": {