johannesbrandenburger/typst-mcp
If you are the rightful owner of typst-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.
Typst MCP Server is an implementation of the Model Context Protocol that facilitates interaction between AI models and Typst, a markup-based typesetting system.
Typst MCP Server
Typst MCP Server is an MCP (Model Context Protocol) implementation that helps AI models interact with Typst, a markup-based typesetting system. The server provides tools for converting between LaTeX and Typst, validating Typst syntax, and generating images from Typst code.
Available Tools
⚠️ Currently all the functionality is implemented as
tools, because Cursor and VS Code are not able to handle the other primitives yet.
The server provides the following tools:
-
list_docs_chapters(): Lists all chapters in the Typst documentation.- Lets the LLM get an overview of the documentation and select a chapter to read.
- The LLM should select the relevant chapter to read based on the task at hand.
-
get_docs_chapter(route): Retrieves a specific chapter from the Typst documentation.- Based on the chapter selected by the LLM, this tool retrieves the content of the chapter.
- Also available as
get_docs_chapters(routes: list)for retrieving multiple chapters at once.
-
latex_snippet_to_typst(latex_snippet): Converts LaTeX code to Typst using Pandoc.- LLMs are better at writing LaTeX than Typst, so this tool helps convert LaTeX code to Typst.
- Also available as
latex_snippets_to_typst(latex_snippets: list)for converting multiple LaTeX snippets at once.
-
check_if_snippet_is_valid_typst_syntax(typst_snippet): Validates Typst code.- Before sending Typst code to the user, the LLM should check if the code is valid.
- Also available as
check_if_snippets_are_valid_typst_syntax(typst_snippets: list)for validating multiple Typst snippets at once.
-
typst_to_image(typst_snippet): Renders Typst code to a PNG image.- Before sending complex Typst illustrations to the user, the LLM should render the code to an image and check if it looks correct.
- Only relevant for multi modal models.
Getting Started
-
Clone this repository
git clone https://github.com/johannesbrandenburger/typst-mcp.git
-
Clone the typst repository
git clone https://github.com/typst/typst.git
-
Run the docs generation in the typst repository
cargo run --package typst-docs -- --assets-dir ../typst-mcp/typst-docs --out-file ../typst-mcp/typst-docs/main.json- Make sure to adjust the path to your local clone of the typst-mcp repository
- This will generate the
main.jsonand the assets in thetypst-docsfolder
-
Install required dependencies:
uv sync(install uv if not already installed) -
Install Typst
Running the Server
Local Installation
Execute the server script:
python server.py
Or install it in Claude Desktop with MCP:
mcp install server.py
Or use the new agent mode in VS Code:
Agent mode: available to all users and supports MCP
Docker
For Docker installation and usage, see for detailed build information and instructions.
Platform Configuration
This MCP server can be integrated with various AI coding platforms. Below are configuration instructions for popular platforms:
Claude Desktop
Add the following to your Claude Desktop configuration file (claude_desktop_config.json):
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
Or for local installation:
{
"mcpServers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
Cursor
Create .cursor/mcp.json in your project root for project-specific configuration:
{
"mcpServers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
Or create ~/.cursor/mcp.json for global configuration. For local installation:
{
"mcpServers": {
"typst": {
"command": "python",
"args": ["/path/to/typst-mcp/server.py"]
}
}
}
Roo Code
Create .roo/mcp.json in your project root for project-specific configuration:
{
"mcpServers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
Or edit global settings via Roo Code MCP settings view. For local installation:
{
"mcpServers": {
"typst": {
"command": "python",
"args": ["/path/to/typst-mcp/server.py"]
}
}
}
OpenCode
Create or edit the opencode.json file in your project root:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"typst": {
"type": "local",
"command": ["docker", "run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
],
"enabled": true
}
}
}
For local installation:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"typst": {
"type": "local",
"command": ["python", "/path/to/typst-mcp/server.py"],
"enabled": true
}
}
}
Claude Code
Configure MCP servers in your Claude Code settings file (~/.claude/settings.json):
{
"mcpServers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
Or for local installation:
{
"mcpServers": {
"typst": {
"command": "python",
"args": ["/path/to/typst-mcp/server.py"]
}
}
}
VS Code with Agent Mode
VS Code's Agent Mode has native MCP support (no extension required):
- Enable Agent Mode in VS Code settings (
chat.agent.enabled) - Create
.vscode/mcp.jsonin your workspace:
{
"servers": {
"typst": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"ghcr.io/johannesbrandenburger/typst-mcp:latest"
]
}
}
}
For global configuration, add to your user profile via MCP: Open User Configuration command.
JSON Schema of the Typst Documentation
⚠️ The schema of the typst documentation is not stable and may change at any time. The schema is generated from the typst source code and is not guaranteed to be complete or correct. If the schema changes, this repository will need to be updated accordingly, so that the docs functionality works again.