drf-llm-gateway

Turn your existing Django REST Framework viewsets into tools an LLM can call — OpenAI function-calling format, MCP (Model Context Protocol) tool format, or plain JSON Schema — generated automatically from the serializers you already have.

from rest_framework import viewsets
from drf_llm_gateway import expose_as_tool


@expose_as_tool(actions=["list", "retrieve", "create"])
class ArticleViewSet(viewsets.ModelViewSet):
    queryset = Article.objects.all()
    serializer_class = ArticleSerializer

from drf_llm_gateway import default_registry
from drf_llm_gateway.openai_tools import to_openai_tools

tools = to_openai_tools(default_registry)
# [{"type": "function", "function": {"name": "article_list", "description": "...",
#   "parameters": {"type": "object", "properties": {...}}}}, ...]

Why

Wiring an LLM agent up to call your existing API usually means hand-writing and hand-maintaining a parallel set of tool/function schemas — which drift out of sync with your serializers the moment either side changes. This package derives the schema from the serializer itself at generation time, so there is nothing to keep in sync: change a field on the serializer, regenerate the tools, done.

Features

• Automatic JSON Schema generation from any DRF serializer — every standard field type, nested serializers, list serializers, related fields, choice fields, and more.
• OpenAI function-calling format and MCP tool format exporters.
• Tool registry — a decorator (@expose_as_tool) that registers a viewset's actions as callable tools without any schema duplication.
• Runtime executor — validates arguments against the real serializer and dispatches to the real viewset action, respecting DRF permission and authentication classes.
• Auth/permission metadata attached to every tool definition, so a gateway can decide what an agent identity may call before it calls it.
• Schema versioning — every generated tool carries a content hash that changes whenever its schema changes, so consumers can detect drift.
• CLI (generate_llm_tools management command) to export the full tool set as JSON for any external agent framework.
• Fully typed, PEP 561 compatible, mypy --strict clean.

Installation

pip install drf-llm-gateway

Requires Python 3.10+, Django 4.2+, and Django REST Framework 3.14+.

Quick Start

See docs/quickstart.md for a complete walkthrough, including calling a tool end-to-end from a validated argument dict.

Documentation

Full documentation: https://mahmoudgshake.github.io/MahmoudPackages/drf-llm-gateway

• Getting Started
• Installation
• Configuration / Settings
• Quick Start
• Advanced Usage
• Architecture
• API Reference
• Examples
• Common Patterns
• Performance
• Security
• Testing
• Deployment
• FAQ
• Troubleshooting

Contributing

Contributions are welcome — see CONTRIBUTING.md.

License

MIT — see LICENSE.