Problem

AI agents can generate text, but working reliably with Microsoft Word files remains difficult: developers must otherwise handle DOCX/XML internals, while direct automation risks silently overwriting documents rather than producing reviewable suggestions. Vespper targets this workflow pain by making agent-generated edits cheaper and more performant while preserving Word’s tracked-change model.

The primary use case is regulated, document-heavy work where human review and an auditable revision trail are essential, especially regulatory, compliance, legal, and life-science submission workflows. In these settings, the cost is not only engineering effort but also review friction and the risk of an incorrect or untraceable change reaching a formal document.

Product / Service

Vespper is a hosted DOCX Model Context Protocol (MCP) server that lets agents read Word documents and apply edits without requiring developers to manipulate XML. Its read_document tool converts a DOCX into an HTML representation for the agent to reason over, while edit_document applies {old, new} HTML edits and returns a new document; edits remain visible as Word tracked changes until a human accepts or rejects them.

The delivery model is an API-key-authenticated endpoint at mcp.vespper.com/mcp, exposed through MCP to tools and agent environments including Cursor, Codex CLI, Gemini CLI, Copilot CLI, Claude Desktop, SDKs, and raw MCP clients. This gives agent builders a reusable document-editing layer with built-in anchoring, batching, and tracked-change semantics, reducing implementation complexity while keeping a human approval step in the workflow.

Market

Vespper competes in AI developer tools, MCP infrastructure, document automation, and vertical software for regulated knowledge work. Its initial wedge is the DOCX interface layer for agents, with a particularly clear fit for regulatory, compliance, legal, medical-writing, and clinical-affairs teams. Adjacent alternatives include open-source or community DOCX MCP servers such as Office-Word-MCP-Server and Docx MCP Server, as well as libraries such as docx-editor that provide programmatic Word editing and tracked changes; conventional Word automation is the broader incumbent alternative.

Vespper is an early-stage, active YC Fall 2024 company founded in 2024, with a reported two-person team and a $500,000 pre-seed round dated October 2024. Its current product has a developer dashboard, API keys, hosted MCP endpoint, and editing playground, while its LinkedIn activity records attendance at a regulatory-industry conference and dozens of potential-customer leads. The available evidence supports early commercial validation and product availability, but does not disclose revenue, paying-customer counts, or enough information to call the company demonstrably revenue-generating; the YC profile also retains older material describing an earlier AI on-call product, so current traction should be treated as not publicly quantified.