FastMCP LaTeX Server (tex-mcp)

1[![MseeP.ai Security Assessment Badge](https://mseep.net/pr/devroopsaha744-texmcp-badge.png)](https://mseep.ai/app/devroopsaha744-texmcp)
2 
3# FastMCP LaTeX Server (tex-mcp)
4 
5A small FastMCP-based Microservice that renders LaTeX to PDF. The server exposes MCP tools
6to render raw LaTeX or templates and produces artifacts (a .tex file and  .pdf)
7under `src/artifacts/`.
8 
9This repository is prepared to run locally and to be loaded by Claude Desktop (via the
10Model Context Protocol). The default entrypoint is `run_server.py`.
11 
12## Demo
13![Demo screenshot](assets/demo-1.png)
14![Demo screenshot](assets/demo-2.png)
15![Demo screenshot](assets/demo-3.png)
16---
17 
18## Quick features
19- Render raw LaTeX to `.tex` and (optionally) `.pdf` using pdflatex
20- Render Jinja2 templates and compile to PDF
21- Designed to run as an MCP server for Claude Desktop and other MCP-capable clients
22 
23Tools exposed by this MCP server
24- Total tools: 5
25  - render_latex_document — write LaTeX and optionally compile to PDF
26  - render_template_document — render a Jinja2 template and optionally compile
27  - list_templates — list available templates
28  - list_artifacts — list files produced in `src/artifacts/`
29  - get_template — return the raw contents of a template file so clients can inspect it before rendering
30---
31 
32## Getting started (local development)
33 
34Prerequisites
35- Python 3.10+ (the project uses modern pydantic/fastapi stack)
36- LaTeX toolchain (pdflatex) for PDF compilation (optional; if missing, server returns .tex only)
37 
381) Create a project virtualenv and install deps
39 
40Clone from GitHub
41 
42If you want to work from the canonical repository on GitHub, clone it first:
43 
44```powershell
45git clone https://github.com/devroopsaha744/TexMCP.git
46cd TexMCP
47```
48 
49After cloning you can follow the venv creation and install steps below.
50 
51 
52```powershell
53python -m venv .venv
54. .\\.venv\\Scripts\\Activate.ps1
55python -m pip install --upgrade pip
56pip install -r requirements.txt
57```
58 
592) Run the server directly (stdio mode - used by Claude Desktop)
60 
61```powershell
62. .\\.venv\\Scripts\\Activate.ps1
63python .\\run_server.py
64# or run the venv python explicitly if you don't activate
65.# .venv\\Scripts\\python.exe run_server.py
66```
67 
68If run in stdio mode the server will speak MCP over stdin/stdout (this is what Claude Desktop
69expects when it spawns the process). If you prefer HTTP, edit `run_server.py` and switch the
70transport to `http` (see commented code) and run via `uv run` or `uvicorn`.
71 
723) Artifacts
73 
74Rendered outputs are placed in `src/artifacts/`. For each job you should see a `.tex` file and
75— if `pdflatex` is available — a matching `.pdf`.
76 
77Templates
78 - Several example templates live in `src/mcp_server/templates/`. There are 15 templates included (for example `sample_invoice.tex.j2`, `sample_letter.tex.j2`, `sample_resume.tex.j2`). Use `list_templates` to get the full list programmatically. The templates are deliberately simple and ready to customize — add your own `.tex.j2` files to that folder to expand the catalog.
79 
80Included templates (in `src/mcp_server/templates/`)
81 
82- `default.tex.j2` (base example template)
83- `sample_invoice.tex.j2`
84- `sample_invoice2.tex.j2`
85- `sample_letter.tex.j2`
86- `sample_report.tex.j2`
87- `sample_resume.tex.j2`
88- `sample_presentation.tex.j2`
89- `sample_certificate.tex.j2`
90- `sample_coverletter.tex.j2`
91- `sample_poster.tex.j2`
92- `sample_thesis.tex.j2`
93- `sample_receipt.tex.j2`
94- `sample_recipe.tex.j2`
95- `sample_poem.tex.j2`
96- `sample_cv.tex.j2`
97 
98---
99 
100## Integration with Claude Desktop (quick)
101 
102Recommended: use the `fastmcp` CLI installer which will set things up to run from the project directory and use the project venv.
103 
104From your project root (with the venv already created and deps installed):
105 
106```powershell
107fastmcp install claude-desktop run_server.py --project C:\\Users\\DEVROOP\\Desktop\\tex-mcp
108```
109 
110This ensures `uv` runs inside the project directory and uses the project's environment. After the installer runs, fully quit and restart Claude Desktop.
111 
112Manual Claude Desktop config
113If you edit Claude's config yourself (Windows: `%APPDATA%\\Claude\\claude_desktop_config.json`), add a single server entry that points to the project Python executable. Example (replace paths if needed):
114 
115```json
116{
117  "mcpServers": {
118    "FastMCP-LaTeX-Server": {
119      "command": "C:\\\\Users\\\\DEVROOP\\\\Desktop\\\\tex-mcp\\\\venv\\\\Scripts\\\\python.exe",
120      "args": [
121        "C:\\\\Users\\\\DEVROOP\\\\Desktop\\\\tex-mcp\\\\run_server.py"
122      ],
123      "cwd": "C:\\\\Users\\\\DEVROOP\\\\Desktop\\\\tex-mcp",
124      "transport": "stdio"
125    }
126  }
127}
128```
129 
130Notes
131- Do NOT point Claude at the virtualenv `activate` script — it is a shell helper and not an executable. Point Claude to the `python.exe` inside the venv (or to `uv.exe` inside the venv if you installed `uv`).
132- After any config changes, fully restart Claude Desktop.
133 
134---
135 
136## Docker
137 
138This project includes a Dockerfile so you can run the MCP server in a container.
139 
140Build (no LaTeX):
141 
142```bash
143docker build -t fastmcp-latex:latest .
144```
145 
146Build with LaTeX (larger image):
147 
148```bash
149docker build --build-arg INSTALL_TEX=1 -t fastmcp-latex:with-tex .
150```
151 
152Run (HTTP mode exposed on port 8000):
153 
154```bash
155docker run -p 8000:8000 --rm --name fastmcp-latex fastmcp-latex:latest
156```
157 
158Notes
159- The container sets `MCP_TRANSPORT=http` by default. Inside the container the server binds to `0.0.0.0:8000`.
160- If you want to run the server in `stdio` mode in a container you can override the env var:
161 
162```bash
163docker run -e MCP_TRANSPORT=stdio ...
164```
165 
166Artifact persistence
167- To persist rendered artifacts on the host, bind mount the `src/artifacts` directory:
168 
169```bash
170docker run -p 8000:8000 -v $(pwd)/src/artifacts:/app/src/artifacts fastmcp-latex:latest
171```
172 
173---
174 
175You can  Use a Model Context Protocol / FastMCP client library (Like OpenAI Responses API) in your agent code to call tools programmatically. For example, in Python you can use the `mcp` or `fastmcp` client (see library docs) to connect to `http://localhost:8000/mcp` and call `render_latex_document` with arguments.
176 
177Security notes
178- If you expose the HTTP endpoint beyond localhost, secure it (TLS, firewall, or authentication) — rendering arbitrary LaTeX can pose risks (shell commands in templates, large resource use).
179 
180---
181 
182## Contributing
183 
184Thanks for wanting to contribute! See `CONTRIBUTING.md` for the development workflow, commit style, and how to open issues and pull requests.
185 
186---
187 
188## License
189 
190This project is released under the MIT License — see `LICENSE`.
191 
192