adeu

1. Purpose Adeu is positioned as a "Track Changes for the LLM era" tool. Large language models are good at handling Markdown, while reviewers are accustomed to using Word's "Track Changes" feature, and Adeu bridges the two. It is a docx ↔ LLM translator: it projects .docx office files as Markdown that is easy to edit, then writes the edits back into the document in the form of OOXML tracked changes (redlines). It is also a Model Context Protocol (MCP) server (offering both Python and Node.js implementations) and an accompanying SDK, essentially a "virtual DOM" for Microsoft Word, providing a bidirectional abstraction layer that lets AI agents freely edit document text without breaking the underlying formatting or the complex DOCX XML.

2. Main capabilities As an intelligent agent, Adeu handles AI edits as safe, atomic transactions, divided into three stages:

3. Read: Translate the document (from disk or live Word) into LLM-friendly CriticMarkup, along with a "semantic appendix" containing defined terms, cross-references, and possible spelling errors, so the agent starts from the semantic structure rather than raw data.

4. Validate: Act as a strict safety gate, automatically blocking ambiguous text matches or invalid structural changes, protecting document integrity before changes touch the file.

5. Apply: Convert the AI's text edits into native Word tracked changes, with Adeu handling the complex XML underneath to ensure existing layout, fonts, and margin comments are fully preserved.

The specific tools available to the agent include:

• read_docx(clean_view=True): Reads the final "clean" version text to understand context, and can use search_query and pagination filtering to locate specific clauses without reading the entire document. This tool supports the MCP Apps UI protocol, and when reading a document it renders a custom interactive Markdown view directly within the chat window.
• process_document_batch: Commit & Negotiate Mode, which can apply a unified change list. The type "modify" is used for specific find-and-replace text edits (supports match_mode="all" with regex=True for batch updates); the types "accept", "reject", and "reply" are used to manage existing revisions and comments by ID.
• finalize_document: Pre-Send Scrub before sending, stripping dangerous metadata, author names, and internal tracking IDs, locking the document (protection_mode="read_only") and preparing it for distribution.

III. How to Use Adeu can be installed into an AI assistant as an MCP server, or used as a Claude Code plugin or Agent Skill, a CLI tool, or a local developer toolchain.

• Claude Code plugin: includes a built-in Agent Skill that teaches Claude how to use the engine effectively. For best results, you can also connect the Node MCP server (npx -y @adeu/mcp-server) or the Python MCP server (uvx --from adeu adeu-server); the plugin also works without an MCP server, falling back to driving the uvx adeu CLI via Bash.
• Other Skills-compatible agents (Cursor, Windsurf, VS Code Copilot, etc.): their redlining skill follows the open Agent Skills specification, can be installed into the agent's skills directory, and activates automatically when asked to redline, edit, or review .docx files.
• Claude Desktop: you can download the Adeu.mcpb file from the GitHub Releases page and install it in Settings > Extensions via the Extension Developer area under Advanced settings by selecting Install Extension....
• Gemini CLI: a native Gemini CLI extension is provided.
• Other MCP clients (Cursor, Windsurf, etc.): can be configured via JSON, using either the Node.js or Python backend. Live MS Word integration on Windows requires the Python backend.
• Smithery package manager: you can install Adeu using Smithery.
• Live MS Word integration: on Windows with Microsoft Word installed, Adeu can act as a live copilot, editing the currently active document in front of you, and requires running the Python MCP server backend.

IV. Developer Tools

• The Python toolchain is managed via uv.
• TypeScript SDK: the entire core parsing and diff engine is also available in pure TypeScript; see the @adeu/core documentation.
• n8n community node (n8n-nodes-adeu): exposes the full engine (extract Markdown, apply revisions, generate diffs, finalize documents) as ready-to-use nodes, suitable for deterministic pipelines and AI Agent tool calls.
• LangChain integration: langchain-adeu is the official integration package, exposing Adeu's local, offline-capable document manipulation tools to the LangChain ecosystem, and can be packaged as tools in agent workflows.
• Ecosystem and integrations: the core is strictly focused on OpenXML safety, while third-party integrations are placed in the ecosystem/ directory, hosting policies and guidelines for contributed legal validation workflows, CLM sync scripts, specialized multi-agent architectures, and more.
• Fidelity notes: exactly what the text projection preserves, what it normalizes (lists, styles, synthesized page breaks), and what is read-only are specified in detail in docs/FIDELITY.md.

V. Use Cases and Limitations Use cases: legal, compliance, or document review workflows that need AI agents to make non-destructive revisions to Word documents; lossless redline editing that is difficult to achieve with standard libraries (such as python-docx) beyond generating documents from scratch; teams that want to connect automation via MCP, CLI, n8n, or LangChain. Privacy and licensing: by default, the core Adeu redlining engine and local file tools are fully open source and run entirely on your machine; Adeu does not send local documents back (though the LLM provider you choose will naturally process the text read by the agent). For teams that need end-to-end workflows, you can connect Adeu Cloud to unlock capabilities such as email processing and retrieval, including providing an extended MCP server that supports securely retrieving email leads, extracting documents, and automatically drafting, so contracts can be handled directly in the inbox. The project uses the MIT license, openOpen source and free for commercial applications. Note: The above content is compiled based on the provided public information. Regarding specific pricing, platform support scope, and other features not mentioned, the information is not yet available.