MCP Server Office

Instructions

Implementation Reference

• mcp_server_office/office.py:78-127 (handler)

  The main handler function for the read_docx tool. It validates the path, loads the DOCX document, processes paragraphs (including images and track changes), extracts table text, and formats the output with paragraph separators.

  async def read_docx(path: str) -> str:
      """Read docx file as text including tables.
      
      Args:
          path: relative path to target docx file
      Returns:
          str: Text representation of the document including tables
      """
      if not await validate_path(path):
          raise ValueError(f"Not a docx file: {path}")
      
      document = Document(path)
      content = []
  
      paragraph_index = 0
      table_index = 0
      
      # 全要素を順番に処理
      for element in document._body._body:
          # パラグラフの処理
          if element.tag == W_P:
              paragraph = document.paragraphs[paragraph_index]
              paragraph_index += 1
              # 画像のチェック
              if paragraph._element.findall(f'.//{W_DRAWING}', WORDML_NS):
                  content.append("[Image]")
              # テキストのチェック
              else:
                  text = process_track_changes(paragraph._element)
                  if text.strip():
                      content.append(text)
                  else:
                      # 空行を抜くと編集時に困るので、空行でも追加
                      content.append("")
          # テーブルの処理
          elif element.tag == W_TBL:
              table = document.tables[table_index]
              table_index += 1
              table_text = extract_table_text(table)
              content.append(f"[Table]\n{table_text}")
  
      separator = [f"--- Paragraph {i} ---" for i in range(len(content))]
      
      result = []
      for i, p in enumerate(content):
          result.append(separator[i])
          result.append(p)
          
      return "\n".join(result)

• mcp_server_office/tools.py:3-22 (schema)

  The Tool object definition providing the schema, name, and description for the read_docx tool, including input schema validation for the 'path' parameter.

  READ_DOCX = types.Tool(
      name="read_docx",
      description=(
          "Read complete contents of a docx file including tables and images."
          "Use this tool when you want to read file endswith '.docx'."
          "Paragraphs are separated with two line breaks."
          "This tool convert images into placeholder [Image]."
          "'--- Paragraph [number] ---' is indicator of each paragraph."
      ),
      inputSchema={
          "type": "object",
                  "properties": {
                  "path": {
                          "type": "string",
                          "description": "Absolute path to target file",
                      }
                  },
          "required": ["path"]
      }
  )

• mcp_server_office/office.py:358-360 (registration)

  Registers the read_docx tool (as READ_DOCX) in the MCP server's list_tools handler.

  @server.list_tools()
  async def list_tools() -> list[types.Tool]:
      return [READ_DOCX, EDIT_DOCX_PARAGRAPH, WRITE_DOCX, EDIT_DOCX_INSERT]

• mcp_server_office/office.py:367-369 (registration)

  Dispatches calls to the read_docx tool handler within the server's call_tool function.

  if name == "read_docx":
      content = await read_docx(arguments["path"])
      return [types.TextContent(type="text", text=content)]

• mcp_server_office/office.py:62-77 (helper)

  Helper function to extract text from paragraphs while handling track changes (insertions). Used in read_docx.

  def process_track_changes(element: OxmlElement) -> str:
      """Process track changes in a paragraph element."""
      text = ""
      for child in element:
          if child.tag == W_R:  # Normal run
              for run_child in child:
                  if run_child.tag == W_T:
                      text += run_child.text if run_child.text else ""
          elif child.tag.endswith('ins'):  # Insertion
              inserted_text = ""
              for run in child.findall('.//w:t', WORDML_NS):
                  inserted_text += run.text if run.text else ""
              if inserted_text:
                  text += inserted_text
      return text