VSCode拡張機能 - ClaudeやCopilot ChatからExcel VBAを直接操作するA VSCode extension leveraging MCP enables AI assistants like Claude and GitHub…

A newly described VSCode extension illustrates how the Model Context Protocol (MCP) can connect general-purpose AI assistants such as Anthropic's Claude and GitHub Copilot Chat directly to Excel's VBA environment. The idea matters because it aims to fold a decades-old but still widely used automation language into the same conversational, AI-assisted workflow that many developers now rely on for modern programming tasks, potentially lowering the friction of maintaining the spreadsheet macros that quietly run inside countless businesses.

At the center of the setup is MCP, an open standard introduced by Anthropic in late 2024 to give AI models a consistent way to reach external tools and data. MCP uses a client-server design: an MCP server exposes a set of capabilities, such as functions to call or resources to read, and an MCP-aware client, which in this case is the AI assistant running inside the editor, can discover and invoke them. Rather than pasting code back and forth, the assistant issues structured requests that a server translates into concrete actions. The pattern has been adopted across a growing range of integrations, from file systems and databases to issue trackers and browsers, and applying it to Office automation is a logical extension of that trajectory.

In practical terms, the extension appears to act as, or work alongside, an MCP server that bridges the AI assistant and the Excel application. Excel's macros are written in Visual Basic for Applications (VBA) and stored inside the workbook's project, which is normally edited through the Visual Basic Editor rather than a text file on disk. Programmatic access to that project typically relies on Windows COM automation and the VBA extensibility interfaces, so a server sitting between the AI and Excel would be responsible for reading existing modules, inserting or replacing procedures, and, where permitted, triggering a macro to run. This is what lets a user describe a task in natural language and have generated VBA land in the correct module instead of an isolated chat window.

The workflow this enables is straightforward to imagine. A user could ask the assistant to write a routine that formats a report, loops through worksheets, or cleans imported data, then have the code inserted and tested without leaving the editor. Because the AI can also read back the current state of the project, it can attempt to refactor existing macros, explain what an inherited procedure does, or adjust code in response to an error message. That closed loop, generating, applying, and observing results, is generally what distinguishes an MCP integration from simply prompting a chatbot for a snippet.

Some context is useful for weighing the approach. VBA remains deeply embedded in finance, operations, and administrative work, yet it lacks the rich tooling of contemporary languages, and finding developers comfortable with it is increasingly difficult. Microsoft has pushed newer automation options, including Office Scripts with TypeScript for Excel on the web and Power Automate for cross-application workflows, but legacy VBA workbooks are unlikely to disappear soon. A bridge that makes those macros easier to read and modify with AI assistance addresses a real maintenance gap, even if it does not modernize the underlying platform.

Several caveats also deserve mention. COM-based automation of Excel is primarily a Windows capability, so the integration is likely tied to that platform. Allowing an AI assistant to insert and execute macros raises familiar security considerations, since VBA can touch the file system and other applications, and organizations often restrict macro execution for exactly that reason. As with any generated code, output should be reviewed and tested rather than trusted outright, and users would need to confirm how the tool handles workbook access and permissions.

More broadly, the project reflects a wider movement to make MCP the connective tissue between AI models and everyday software. GitHub Copilot Chat and Claude both support MCP-style tool use, and the ecosystem of community-built servers has expanded quickly. Whether this particular extension becomes widely adopted is uncertain, but it is a concrete example of AI tooling reaching into an environment, Office automation, that has largely sat outside the modern developer stack.