precis-mcp 0.4.0

MCP server for compressed structured context on .docx and .tex documents

precis-mcp

A Model Context Protocol (MCP) server that gives LLMs compressed, structured context for .docx and .tex documents. It maintains a heading tree with RAKE keyword summaries so the LLM can navigate and edit documents without flooding context.

Features

• 5 tools — activate, toc, get, put, move
• Dual addressing — 5-char content-hash slugs + positional heading paths (S1.2 = heading, S1.2¶3 = paragraph)
• RAKE keyword extraction — stateless, zero-dependency precis generation (<5ms)
• DOCX citations — [@key] round-trip with styled hyperlinks and bibliography entries
• Heading safety — # | Title pipe separator + auto-strip of LLM-inserted section numbers
• Track changes — put() writes Word revision markup by default
• LaTeX support — \input/\include resolution, \label{} aliases, .bib parsing
• Atomic I/O — every call reads fresh from disk, no stale state

Installation

uv pip install -e ".[dev]"

Usage

Run as an MCP server:

precis

Or use with an MCP client:

from precis.tools import Session, activate, toc, get, put

session = Session()
await activate(session, "paper.docx")
await toc(session)
await get(session, id="KR8M2")
await put(session, id="KR8M2", text="Updated text.", mode="replace")

Configuration

~/.config/precis/precis.toml:

[precis]
author = "precis"   # track-changes author name (DOCX only)

Testing

uv run python -m pytest tests/ -v

License

GPL-3.0-or-later — see LICENSE.