TexMCP

TexMCP is a FastMCP-based LaTeX rendering server. It exposes five MCP tools to clients: render_latex_document writes LaTeX code to a .tex file and can optionally compile it to PDF; render_template_document renders a Jinja2 template into a .tex file and can optionally compile to PDF; list_templates lists available templates from src/mcp_server/templates; list_artifacts lists the generated files in src/artifacts; and get_template returns the raw contents of a template file for inspection. By default the server communicates over stdio suitable for Claude Desktop, but you can switch to HTTP transport by adjusting the run_server.py configuration and using uvicorn or similar. Clients connect to the MCP endpoint and invoke these tools with structured arguments to produce and retrieve LaTeX artifacts.

Prerequisites:

• Python 3.10+
• LaTeX toolchain (pdflatex) if you want PDF output
• Git (optional, for cloning)

1. Clone the repository:

git clone https://github.com/devroopsaha744/TexMCP.git
cd TexMCP

2. Create and activate a virtual environment:

# Windows
python -m venv .venv
.\.venv\Scripts\Activate.ps1

# macOS/Linux
python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Run the server (stdio mode by default):

# Windows
.\.venv\Scripts\Activate.ps1
python .\run_server.py

# If you prefer to run explicitly with the venv Python
.\.venv\Scripts\python.exe run_server.py

5. Optional — run HTTP server instead of stdio (edit run_server.py to switch transport to http) and run with uvicorn if desired.

6. Verify artifacts directory:

ls -la src/artifacts

Notes and tips:

• Rendered outputs appear in src/artifacts/. Each job yields a .tex file and, if pdflatex is available, a corresponding .pdf.
• Templates reside in src/mcp_server/templates/. Use list_templates to enumerate them programmatically or inspect the files directly.
• If you plan to expose the HTTP endpoint, ensure proper security (TLS, firewall, or authentication) since LaTeX rendering can be resource-intensive and may pose risks when running arbitrary templates.
• For Claude Desktop integration, you can use the recommended fastmcp installer to automate environment alignment, or configure Claude Desktop to point at the local Python executable and run_server.py as described.
• When running via Docker, MCP_TRANSPORT can be set to stdio or http, and you can persist artifacts by mounting src/artifacts as a volume.