Awesome MCP › Software Development

johannesbrandenburger/typst-mcp

⭐ 177 Python repository created 2025-04-18

typst-mcp is a Python MCP server that helps AI models write correct Typst, the markup-based typesetting system. It addresses a concrete weakness: models have seen far more LaTeX than Typst, so the server both supplies authoritative documentation and gives the model a way to check its output before showing it to a user. Five tools are provided, all exposed as tools rather than as other MCP primitives because, as the README notes, Cursor and VS Code could not yet handle those. list_docs_chapters returns the chapter list of the Typst documentation so the model can choose what is relevant to the task at hand. get_docs_chapter fetches a chapter by route, with a batch variant taking several routes at once. latex_snippet_to_typst converts a LaTeX snippet to Typst using Pandoc, again with a batch variant, letting the model draft in the notation it knows best and translate afterwards. check_if_snippet_is_valid_typst_syntax validates a snippet, also available in batch form, so invalid code is caught and corrected before delivery. typst_to_image renders a snippet to a PNG, which is useful for multimodal models that can look at a complex illustration and judge whether it came out right. Setup requires cloning both this repository and the Typst repository, running the typst-docs package to generate a main.json documentation dump and its assets into the server's typst-docs folder, installing dependencies with uv, and installing Typst itself. The server can be run directly as a script, installed into Claude Desktop, or used from a prebuilt container image published to the GitHub Container Registry; the README gives configuration snippets for Claude Desktop, Claude Code, Cursor, Roo Code, OpenCode and VS Code agent mode. The documentation JSON schema is generated from Typst sources and is not guaranteed to stay stable.

https://github.com/johannesbrandenburger/typst-mcp

mcpmcp-servertypstlatextypesettingdocumentationpandocpython