Tool reference

This page describes the public chart, table, and asset tools. Call tools/list on the MCP endpoint you use to retrieve its available tools and current input schemas.

Pass the exact name from that response as params.name in tools/call. Use the tool's inputSchema for the complete set of fields and supported values. Refresh cached definitions when you reconnect or change endpoints.

#Chart tools

Each chart tool accepts a chart definition and an optional assetId:

• Omit assetId to create an asset.
• Include assetId to update the existing asset.
• Pass rules and annotations as part of the chart definition.

#Example: create a column chart

After connecting and authenticating, send this JSON-RPC body to your MCP endpoint with method set to tools/call:

{
  "jsonrpc": "2.0",
  "id": "2",
  "method": "tools/call",
  "params": {
    "name": "create_or_update_bar_chart",
    "arguments": {
      "assetName": "Quarterly revenue",
      "chart": {
        "data": [
          ["", "Q1", "Q2"],
          ["Revenue", 120, 150]
        ],
        "orientation": "vertical",
        "stackingMode": "clustered"
      }
    }
  }
}

The first row names the categories. Each following row contains a series name and its values. For a new bar or column chart, supply data, orientation, and stackingMode. To update this chart later, add the returned assetId to arguments alongside chart.

#Table tools

The create_or_update_table tool accepts a table definition with required cells and optional settings. It supports formatting, conditional rules, merged cells, row and column sizing, icons, gauges, shapes, flags, images, logos, and in-cell charts. Include an assetId to update an existing table.

Use search_visual_assets before adding built-in icons or flags. It accepts a query, an asset type of all, icon, or flag, and a result limit from 1 to 50.

#Asset tools

Use inspect_object before an exact update when you need the current data values, indexes, settings, or rule configuration.

#Share an asset

Call create_asset_link with assetId and an optional permission:

• view (default) allows access to the rendered image, SVG, and underlying data.
• edit grants edit access and allows the Grunt PowerPoint add-in to load the complete model for local editing. It does not convert the asset into a built-in PowerPoint chart or automatically publish deck edits to the cloud.

The tool returns assetId, latestCommit, linkId, permission, and assetUrl in structuredContent. The assetUrl returns a JSON manifest. Follow links.svg.href for the image, links.export.href for object export, or links.data.href for data. Edit manifests also include links.project.href for the complete Grunt model. Use these returned URLs instead of constructing delivery URLs yourself.

When embedding an asset in PowerPoint, create one fresh edit link per asset per deck. Reuse it for repeated placements within that deck.

#Common response fields

Chart and table creation or update tools return:

• structuredContent.assetId
• structuredContent.assetCommit
• structuredContent.exportGuideUrl
• structuredContent.warnings (when present)

They also return a PNG preview of the rendered asset.

When you create an asset, the tool also attempts to create a permanent view link. On success, structuredContent includes assetUrl, linkId, and permission. If link creation fails, it includes linkError; the asset has still been created. Updating an existing asset returns its new assetCommit without creating another link.

fetch_latest_asset returns the current assetId, assetCommit, exportGuideUrl, and preview.

inspect_object returns the inspected components and, when available, workbook data. get_project_changes returns a structured description of changes since the supplied assetCommit.

#Input validation rules

• Tool schemas reject unknown properties
• Typed settings must use a supported discriminator, such as annotationType, ruleType, or axisType
• Invalid schemas return MCP tool errors

#Handle tool errors

Check both JSON-RPC errors and the tool result's isError field. An HTTP success response does not by itself mean the tool succeeded.

If a call reports an unknown tool, compare params.name with the current tools/list response from the same endpoint. Unknown-tool failures can also be reported as an internal error with a reference ID. See troubleshooting before changing permissions or retrying the request.