Skip to main content
Vespper DOCX MCP lets agents read .docx files as HTML and apply edits as Word tracked changes (suggestions) — never silent overwrites. You connect over MCP from your own agent code. The tools ship the editing instructions agents need — block anchoring, lazy classes, batching, tracked-change semantics — so you do not have to replicate them in a system prompt.

​
The three primitives

Read document

Return a range of pages as HTML.

Search document

Regex-match blocks without loading the whole document.

Edit document

Apply old → new edits as Word tracked changes.
They are stateless and designed to be chained inside an agent loop: search or read to get an anchor, edit to apply it, then thread the returned document into the next call.

​
MCP server

The document is not a tool argument: your client attaches it to each call’s MCP _meta object, so the model sees clean schemas and never spends context on base64. That also means the integration lives in your code — see the Quickstart for framework and native paths.

​
For coding agents

  • Quickstart — framework (OpenAI Agents SDK, Mastra, Vercel AI SDK) and native loops
  • Primitives — parameters and return shapes for the three tools
  • MCP setup — endpoint, tool schemas, and the _meta document channel
  • llms.txt — machine-readable index