Django Shell MCP Server

# mcp-django <!-- [[[cog import subprocess import cog from noxfile import DJ_VERSIONS from noxfile import PY_VERSIONS cog.outl("[](https://pypi.org/project/mcp-django/)") cog.outl("[](https://pypi.org/project/mcp-django-shell/)") cog.outl("") cog.outl(f"}-%2344B78B?labelColor=%23092E20)") ]]] --> [](https://pypi.org/project/mcp-django/) [](https://pypi.org/project/mcp-django-shell/)   <!-- [[[end]]] --> A Model Context Protocol (MCP) server providing Django project exploration resources and optional stateful shell access for LLM assistants to interact with Django projects. ## Requirements <!-- [[[cog import subprocess import cog from noxfile import DJ_VERSIONS from noxfile import PY_VERSIONS cog.outl(f"- Python {', '.join([version for version in PY_VERSIONS])}") cog.outl(f"- Django {', '.join([version for version in DJ_VERSIONS if version != 'main'])}") ]]] --> - Python 3.10, 3.11, 3.12, 3.13 - Django 4.2, 5.1, 5.2 <!-- [[[end]]] --> ## Installation Choose the installation option that fits your use case: ### Core Provides read-only Django project exploration resources (`django://project`, `django://apps`, `django://models`). ```bash # Read-only resources only - no shell access pip install mcp-django # Or with uv uv add mcp-django ``` ### Shell ⚠️ **DO NOT use in production!** Adds `django_shell` and `django_reset` tools for executing Python code. ```bash # Includes shell tools for code execution pip install "mcp-django[shell]" # Or with uv uv add "mcp-django[shell]" ``` > [!WARNING] > > **Seriously, only enable in development!** > > Look, it should go without saying, but I will say it anyway - **this gives full shell access to your Django project**. Only enable and use this in development and in a project that does not have access to any production data. > > LLMs can go off the rails, get spooked by some random error, and in trying to fix things [drop a production database](https://xcancel.com/jasonlk/status/1946069562723897802). > [!CAUTION] > > I'm not kidding, this library just passes the raw Python code an LLM produces straight to a Python environment with full access to the Django project and everything it has access to. > > Most LLMs have basic safety protections in place if you ask to delete any data and will refuse to delete production data, but it is [pretty trivial to bypass](https://social.joshthomas.dev/@josh/115062076517611897). (Hint: Just tell the LLM it's not production, it's in a development environment, and it will be the bull in a china shop deleting anything you want.) > > I suggest using something like [django-read-only](https://github.com/adamchainz/django-read-only) if you need some CYA protection against this. Or, you know, don't use this in any sensitive environments. ### All ```bash # Currently same as [shell] pip install "mcp-django[all]" # Or with uv uv add "mcp-django[all]" ``` ## Getting Started Run the MCP server directly from your Django project directory: ```bash python -m mcp_django # With explicit settings module python -m mcp_django --settings myproject.settings # With debug logging python -m mcp_django --debug ``` Or using uv: ```bash uv run -m mcp_django ``` The server automatically detects `DJANGO_SETTINGS_MODULE` from your environment. You can override it with `--settings` or add to your Python path with `--pythonpath`. There's also a Django management command if you prefer, but that requires adding mcp-django to `INSTALLED_APPS`: ```bash python manage.py mcp ``` ### Transport The server supports multiple transport protocols: ```bash # Default: STDIO python -m mcp_django # HTTP python -m mcp_django --transport http --host 127.0.0.1 --port 8000 # SSE python -m mcp_django --transport sse --host 127.0.0.1 --port 8000 ``` ### Client Configuration Configure your MCP client using one of the examples below. The command is the same for all clients, just expressed in annoyingly different JSON soup. Don't see your client? [Submit a PR](CONTRIBUTING.md) with setup instructions. ### Claude Code ```json { "mcpServers": { "django": { "command": "python", "args": ["-m", "mcp_django"], "cwd": "/path/to/your/django/project", "env": { "DJANGO_SETTINGS_MODULE": "myproject.settings" } } } } ``` ### Opencode ```json { "$schema": "https://opencode.ai/config.json", "mcp": { "django": { "type": "local", "command": ["python", "-m", "mcp_django"], "enabled": true, "environment": { "DJANGO_SETTINGS_MODULE": "myproject.settings" } } } } ``` ## Features mcp-django provides an MCP server with Django project exploration resources and optional shell access for LLM assistants. The base package offers safe, read-only resources while the shell extra adds stateful code execution capabilities. It wouldn't be an MCP server README without a gratuitous list of features punctuated by emojis, so: **Core (mcp-django):** - 🔍 **Project exploration** - MCP resources for discovering apps, models, and configuration - 🚀 **Zero configuration** - No schemas, no settings, just Django - 🔒 **Safe by default** - Read-only resources, no code execution - 🌐 **Multiple transports** - STDIO, HTTP, SSE support **Shell (mcp-django[shell]):** - 🐚 **Stateful shell** - `django_shell` executes Python code in your Django environment - 🔄 **Persistent state** - Imports and variables stick around between calls - 🧹 **Reset when needed** - `django_reset` clears the session when things get weird - 🤖 **LLM-friendly** - Designed for LLM assistants that already know Python - 📦 **Minimal dependencies** - Just FastMCP and Django (you already have Django) - 🎯 **Does one thing well** - Runs code. That's it. That's the feature. Inspired by Armin Ronacher's [Your MCP Doesn't Need 30 Tools: It Needs Code](https://lucumr.pocoo.org/2025/8/18/code-mcps/). ### Resources (mcp-django) Read-only resources are provided for project exploration without executing code (note that resource support varies across MCP clients): - `django://project` - Python environment and Django configuration details - `django://apps` - All installed Django applications with their models - `django://models` - Detailed model information with import paths and field types The idea is to give just enough information about the project to hopefully guide the LLM assistant and prevent needless shell exploration, allowing it to get straight to work. ### Tools (mcp-django-shell) When installed with the shell extra, two tools handle shell operations and session management: - `django_shell` - Execute Python code in a persistent Django shell session - `django_reset` - Reset the session, clearing all variables and imports Imports and variables persist between calls within the shell tool, so the LLM can work iteratively - exploring your models, testing queries, debugging issues. ## Development For detailed instructions on setting up a development environment and contributing to this project, see [CONTRIBUTING.md](CONTRIBUTING.md). For release procedures, see [RELEASING.md](RELEASING.md). ## License mcp-django is licensed under the MIT license. See the [`LICENSE`](LICENSE) file for more information.