@tiwater/office-mcp 0.20.1 → 0.20.3

package/office/contracts/tiwater-office-provider-contract-manifest-v1.json

@@ -2,7 +2,7 @@
   "schema": "tiwater.office-provider-contract-manifest/v1",
   "provider": {
     "id": "@tiwater/office-mcp",
-    "version": "0.20.1"
+    "version": "0.20.3"
   },
   "tools": [
     {

package/office/index.mjs

@@ -194,6 +194,8 @@ const docxNestedObjectIdentity = docxObjectIdentity.pick({
 }).extend({
   gridSpan: z.number().int().positive().optional(),
   verticalMerge: z.string().optional(),
+  verticalMergeOwner: docxAddress.optional(),
+  logicalText: z.string().optional(),
 }).strict();
 const docxObservationNode = z.lazy(() => z.object({
   object: docxNestedObjectIdentity,
@@ -329,7 +331,7 @@ const tools = [
   },
   {
     name: 'docx_read_object',
-    description: 'Read selected rows, cells, or paragraphs from one native DOCX and retain the complete result at output. The response returns compact requested descendants when bounded; if receipt.narrowingRequired is true, request fewer observed addresses or descendants. Use docx_read_table for a table.',
+    description: 'Read selected rows, cells, or paragraphs from one native DOCX and retain the complete result at output. A selected cell exposes its vertical-merge owner and logical text, so a continue cell keeps its physical identity while resolving the restart cell value. The response returns compact requested descendants when bounded; if receipt.narrowingRequired is true, request fewer observed addresses or descendants. Use docx_read_table for a table.',
     inputSchema: inputContract('docx_read_object'),
     outputSchema: docxReadObjectOutput,
     annotations: { readOnlyHint: true, idempotentHint: true },
@@ -337,7 +339,7 @@ const tools = [
   },
   {
     name: 'docx_read_table',
-    description: 'Read exactly one table selected from docx_table_index by native OpenXML address. Each call retains full paragraph and text-node detail for exactly the returned row page at output; it never builds another whole-table data object. The machine response is the compact form of that page: each row and cell keeps its reusable native address, grid position and span, vertical-merge owner, physical text, and logical text. In a vertical merge, restart begins one logical cell and a continue cell is not an independent row value: logicalText resolves the restart cell value while text remains the physical cell value. Process the current page before continuing once with receipt.nextOffset; never revisit an earlier offset. Use docx_read_object only when one selected object needs a narrower descendant view. The provider reports physical structure only; the Agent decides the template and business meaning.',
+    description: 'Read exactly one table selected from docx_table_index by native OpenXML address. Each call retains full paragraph and text-node detail for exactly the returned row page at output; it never builds another whole-table data object. The machine response is the compact form of that page: each row and cell keeps its reusable native address, grid position and span, vertical-merge owner, physical text, and logical text. In a vertical merge, restart begins one logical cell and a continue cell is not an independent row value: logicalText resolves the restart cell value while text remains the physical cell value. A table observation is complete only when receipt.remaining is 0; process each page once in ascending receipt.nextOffset order. Use docx_read_object only when one selected object needs a narrower descendant view. The provider reports physical structure only; the Agent decides the template and business meaning.',
     inputSchema: inputContract('docx_read_table'),
     outputSchema: docxTableReadOutput,
     annotations: { readOnlyHint: true, idempotentHint: true },
@@ -345,7 +347,7 @@ const tools = [
   },
   {
     name: 'docx_copy_content',
-    description: 'Replace content in existing plain-text target paragraphs or cells while retaining target container formatting and source inline meaning. A selection copies a whole object by OpenXML address, or an exact substring when range is attached to a run or text address.',
+    description: 'Replace content in existing plain-text target paragraphs or cells while retaining target container formatting. A whole-object selection copies only that object\'s inline content; a range copies an exact substring from a run or text address. Source table, row, cell, span, and merge structure are not copied.',
     inputSchema: inputContract('docx_copy_content'),
     outputSchema: fixedEditOutput('docx_copy_content'),
     handler: args => fixedEdit('docx_copy_content', args),
@@ -627,6 +629,8 @@ function compactDocxObservation(observation) {
       textPreview: children.length === 0 ? identity.textPreview : null,
       ...(identity.gridSpan === null ? {} : { gridSpan: identity.gridSpan }),
       ...(identity.verticalMerge === null ? {} : { verticalMerge: identity.verticalMerge }),
+      ...(node.object.verticalMergeOwner === null ? {} : { verticalMergeOwner: node.object.verticalMergeOwner }),
+      ...(node.object.logicalText === null ? {} : { logicalText: node.object.logicalText }),
     };
     return {
       object,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiwater/office-mcp",
-  "version": "0.20.1",
+  "version": "0.20.3",
   "description": "Published MCP server for Tiwater Office document capabilities",
   "type": "module",
   "license": "MIT",