Vespper

Developers building AI agents that need to work with .docx files today rely on generic tools like the DOCX skill or Office CLI MCP servers, which are slow, expensive and prone to breaking document styling and formatting. These alternatives are measured by Vespper's own benchmarks as costing more per task, taking longer, and producing less accurate edits. Teams building agentic document workflows, such as filling NDAs with data or applying tracked changes, need a way to let an LLM manipulate Word files without losing formatting fidelity or paying excessive inference and latency costs.

Vespper exposes three tools through an MCP server that let AI agents read, search and edit Microsoft Word documents. Agents work with HTML representations of the document while Vespper applies the corresponding edits to the original DOCX file, preserving styles and formatting and recording changes as native Word tracked changes. It ships SDKs in TypeScript and Python that let developers open a session on a .docx file, connect it to an MCP client, and pass the tools to an LLM agent to execute editing tasks like filling forms from a CSV. The company also publishes open-source examples, including a Word add-in, an ONLYOFFICE plugin, and a browser-based docx editor, showing how to integrate agentic editing with suggestion cards or tracked changes.