django-drf-mcp

django-drf-mcp

Automatically exposes Django REST Framework endpoints as MCP tools with zero configuration, supporting multiple transports and authentication.

Category
Visit Server

README

django-drf-mcp

Add MCP (Model Context Protocol) to any Django REST Framework project. Auto-discovers all DRF endpoints and exposes them as MCP tools — zero boilerplate.

Features

  • Zero config — add "django_drf_mcp" to INSTALLED_APPS, include the URLs, done
  • Auto-discovery — every DRF ViewSet/APIView becomes an MCP tool automatically
  • Endpoint filtering — include or exclude specific endpoints with METHOD:PATH glob patterns
  • Multiple transports — STDIO, SSE, Streamable HTTP, or embedded Django view
  • Authentication — pass auth headers (Token, JWT, Basic, API key) to MCP tool calls via HEADERS setting
  • DRF-integrated MCP view — the /mcp/ endpoint is a DRF APIView, inheriting authentication and permissions from REST_FRAMEWORK settings
  • MCP Docs UI — interactive Swagger-style docs for your MCP tools at /mcp/docs
  • DRF Swagger UI — optional built-in Swagger UI and OpenAPI schema endpoints
  • Fully configurable — control every feature via a single DJANGO_MCP settings dict
  • Auto-configures drf-spectacular — no manual setup needed

Installation

pip install django-drf-mcp

Dependencies (installed automatically)

Quick Start

1. Add to INSTALLED_APPS

# settings.py
INSTALLED_APPS = [
    ...
    "rest_framework",
    "django_drf_mcp",
]

drf-spectacular is auto-injected — you don't need to add it yourself.

2. Include URLs

# urls.py
from django.urls import path, include

urlpatterns = [
    ...
    path("", include("django_drf_mcp.urls")),
]

That's it. The MCP endpoint is now live at /mcp/.

3. Enable MCP Docs UI (requires ASGI)

For interactive documentation of your MCP tools, replace your asgi.py:

# asgi.py
import os
import django

os.environ.setdefault("DJANGO_SETTINGS_MODULE", "myproject.settings")
django.setup()

from django_drf_mcp.asgi import get_asgi_application

application = get_asgi_application()

Run with an ASGI server:

uvicorn myproject.asgi:application --host 0.0.0.0 --port 8000

MCP Docs will be available at /mcp/docs.

Transports

STDIO (Claude Code / Claude Desktop)

python manage.py runmcp --transport stdio

SSE

python manage.py runmcp --transport sse --host 0.0.0.0 --port 8001

Streamable HTTP

python manage.py runmcp --transport streamable-http --host 0.0.0.0 --port 8001

Embedded Django View

The MCP protocol is also served at /mcp/ as a DRF APIView — no separate process needed. Supports GET (health check) and POST (JSON-RPC). Inherits authentication and permission classes from REST_FRAMEWORK settings.

MCP Client Configuration

STDIO

{
  "mcpServers": {
    "my-app": {
      "command": "python",
      "args": ["manage.py", "runmcp", "--transport", "stdio"]
    }
  }
}

With a remote Django server:

{
  "mcpServers": {
    "my-app": {
      "command": "python",
      "args": ["manage.py", "runmcp", "--transport", "stdio", "--base-url", "https://my-app.example.com"]
    }
  }
}

Streamable HTTP — Embedded (same port as Django)

The /mcp/ Django view serves MCP on the same port as your app (e.g. 8000):

{
  "mcpServers": {
    "my-app": {
      "type": "streamable-http",
      "url": "http://localhost:8000/mcp/"
    }
  }
}

With authentication:

{
  "mcpServers": {
    "my-app": {
      "type": "streamable-http",
      "url": "http://localhost:8000/mcp/",
      "headers": {
        "Authorization": "Token abc123..."
      }
    }
  }
}

Streamable HTTP — Standalone (separate port)

runmcp --transport streamable-http starts a standalone FastMCP server (e.g. port 8001) that proxies tool calls to Django on port 8000:

{
  "mcpServers": {
    "my-app": {
      "type": "streamable-http",
      "url": "http://localhost:8001/mcp"
    }
  }
}

With authentication:

{
  "mcpServers": {
    "my-app": {
      "type": "streamable-http",
      "url": "http://localhost:8001/mcp",
      "headers": {
        "Authorization": "Token abc123..."
      }
    }
  }
}

SSE — Standalone (separate port)

runmcp --transport sse starts a standalone FastMCP server (e.g. port 8001) that proxies tool calls to Django on port 8000:

{
  "mcpServers": {
    "my-app": {
      "type": "sse",
      "url": "http://localhost:8001/sse"
    }
  }
}

With authentication:

{
  "mcpServers": {
    "my-app": {
      "type": "sse",
      "url": "http://localhost:8001/sse",
      "headers": {
        "Authorization": "Token abc123..."
      }
    }
  }
}

Note: The client headers authenticate the MCP client to the MCP endpoint. The server-side DJANGO_MCP["HEADERS"] setting (see Authentication) authenticates MCP tool calls to the Django API. When using DRF authentication, configure both for full security.

Settings

All configuration is optional. Add a DJANGO_MCP dict to your settings.py:

DJANGO_MCP = {
    # --- Server ---
    "NAME": "django-drf-mcp",              # MCP server name
    "BASE_URL": "http://localhost:8000",    # Django server URL
    "MCP_PATH": "/mcp/",                   # MCP endpoint path

    # --- Endpoint Filtering ---
    "INCLUDE": [],                          # Only expose matching METHOD:PATH patterns
    "EXCLUDE": [],                          # Remove matching METHOD:PATH patterns

    # --- Authentication ---
    "HEADERS": {},                          # HTTP headers sent with every MCP tool call
    "INTERNAL_TOKEN": "",                   # Shared secret for MCP internal requests (optional)

    # --- MCP Docs UI ---
    "MCP_DOCS_ENABLED": True,              # Enable/disable /mcp/docs
    "MCP_DOCS_TITLE": None,                # Docs page title (defaults to NAME)
    "MCP_DOCS_DESCRIPTION": None,          # Docs description (defaults to OpenAPI description)
    "MCP_DOCS_VERSION": None,              # Docs version (defaults to OpenAPI version)
    "MCP_DOCS_EMOJI": None,                # Emoji shown before the title
    "MCP_DOCS_LINKS": [],                  # Extra links: [{"text": "...", "url": "..."}]
    "MCP_DOCS_ENABLE_CORS": True,          # Enable CORS on docs routes
    "MCP_DOCS_VERBOSE": True,              # Show tools load during server startup

    # --- DRF Swagger / Schema ---
    "SWAGGER_ENABLED": False,              # Enable Swagger UI + OpenAPI schema endpoints
    "SCHEMA_PATH": "/api/schema/",         # Schema endpoint path
    "SWAGGER_PATH": "/api/docs/",          # Swagger UI path
}

Settings Reference

Setting Type Default Description
NAME str "django-drf-mcp" MCP server name, shown in health check and docs
BASE_URL str "http://localhost:8000" URL of the running Django server. Used as the base URL for proxying MCP tool calls and injected into the OpenAPI schema servers list
MCP_PATH str "/mcp/" Path where the MCP endpoint is mounted
INCLUDE list [] Only expose endpoints matching these METHOD:PATH glob patterns. Empty = include all
EXCLUDE list [] Remove endpoints matching these METHOD:PATH glob patterns. Applied after INCLUDE
HEADERS dict {} HTTP headers sent with every MCP tool call (e.g. {"Authorization": "Token ..."})
INTERNAL_TOKEN str "" Shared secret for MCP internal requests. If empty, a per-process auto-generated token is used. Set this for multi-worker or standalone runmcp deployments
MCP_DOCS_ENABLED bool True Enable the MCP Docs UI (Starlette-based, requires ASGI)
MCP_DOCS_TITLE str|None None Docs page title. Falls back to NAME
MCP_DOCS_DESCRIPTION str|None None Docs description. Falls back to the OpenAPI schema description
MCP_DOCS_VERSION str|None None Docs version string. Falls back to the OpenAPI schema version
MCP_DOCS_EMOJI str|None None Emoji displayed before the docs title
MCP_DOCS_LINKS list [] Extra links shown in docs. Each item: {"text": "...", "url": "..."}
MCP_DOCS_ENABLE_CORS bool True Enable CORS on the MCP Docs routes
MCP_DOCS_VERBOSE bool True Show tools load during server startup
SWAGGER_ENABLED bool False Enable the DRF Swagger UI and OpenAPI schema endpoints
SCHEMA_PATH str "/api/schema/" Path for the OpenAPI schema endpoint
SWAGGER_PATH str "/api/docs/" Path for the Swagger UI endpoint

Example: Full Configuration

DJANGO_MCP = {
    "NAME": "products-api",
    "BASE_URL": "http://localhost:8000",
    "MCP_PATH": "/mcp/",
    "HEADERS": {
        "Authorization": "Token abc123...",
    },
    "MCP_DOCS_ENABLED": True,
    "MCP_DOCS_TITLE": "Products MCP Tools",
    "MCP_DOCS_DESCRIPTION": "MCP tools for the Products API",
    "MCP_DOCS_EMOJI": "\U0001f6d2",
    "MCP_DOCS_LINKS": [
        {"text": "Admin", "url": "/admin/"},
        {"text": "GitHub", "url": "https://github.com/myorg/myrepo"},
    ],
    "MCP_DOCS_ENABLE_CORS": True,
    "MCP_DOCS_VERBOSE": False,
    "SWAGGER_ENABLED": True,
}

Example: Minimal (disable docs)

DJANGO_MCP = {
    "NAME": "my-api",
    "BASE_URL": "https://my-api.example.com",
    "MCP_DOCS_ENABLED": False,
}

Endpoint Filtering

Control which DRF endpoints are exposed as MCP tools using INCLUDE and EXCLUDE patterns.

Patterns use the format "METHOD:PATH" with glob-style wildcards (*):

# Only expose GET endpoints under /api/
DJANGO_MCP = {
    "INCLUDE": ["GET:/api/*"],
}

# Expose everything except DELETE operations
DJANGO_MCP = {
    "EXCLUDE": ["DELETE:*"],
}

# Expose /api/ endpoints, but exclude DELETE on users
DJANGO_MCP = {
    "INCLUDE": ["*:/api/*"],
    "EXCLUDE": ["DELETE:/api/users/*"],
}
  • METHOD — HTTP method (GET, POST, PUT, PATCH, DELETE) or * for any method
  • PATH — URL path with * wildcard (e.g. /api/*, /api/users/{id}/)
  • Method matching is case-insensitive
  • If INCLUDE is empty (default), all endpoints are included
  • If EXCLUDE is empty (default), nothing is excluded
  • When both are set, INCLUDE is applied first, then EXCLUDE removes from the result

Authentication

If your DRF endpoints require authentication, there are two layers to configure:

1. Server-side: MCP tool calls to the Django API

Configure the HEADERS setting so MCP tool HTTP calls include credentials:

# DRF TokenAuthentication
DJANGO_MCP = {
    "HEADERS": {
        "Authorization": "Token abc123...",
    },
}

# JWT (e.g. SimpleJWT)
DJANGO_MCP = {
    "HEADERS": {
        "Authorization": "Bearer eyJ...",
    },
}

# Basic Auth
DJANGO_MCP = {
    "HEADERS": {
        "Authorization": "Basic dXNlcjpwYXNz",
    },
}

# Custom API key header
DJANGO_MCP = {
    "HEADERS": {
        "X-API-Key": "my-secret-key",
    },
}

These headers are passed to the underlying httpx.AsyncClient and included in every HTTP request made by MCP tools to your Django API.

SessionAuthentication & MCP Internal Requests

SessionAuthentication uses cookies and CSRF tokens — neither of which the MCP internal proxy has. If your views use SessionAuthentication, MCP tool calls will get 403 Forbidden.

django-drf-mcp solves this with a built-in shared secret. The library automatically generates a per-process cryptographic token and injects it into every internal MCP proxy request via the X-MCP-Internal-Token header.

To use this feature, you must add the IsMCPInternalRequest permission class to any view you want MCP tools to access. Without it, the token header is ignored and normal authentication applies.

Add it to your views using the | (OR) operator:

from rest_framework.permissions import IsAuthenticated
from django_drf_mcp.permissions import IsMCPInternalRequest

class MyViewSet(viewsets.ModelViewSet):
    permission_classes = [IsAuthenticated | IsMCPInternalRequest]

This means:

  • Browser users — must be authenticated via session (or any other auth method)
  • MCP tool calls — bypass session auth via the auto-injected secret token

The token is 256-bit cryptographically random and regenerated on every server restart. External clients cannot guess it.

Multi-worker & standalone runmcp deployments

The auto-generated token is per-process. If the MCP proxy and your DRF views run in different processes (e.g., runmcp standalone mode, or gunicorn with multiple workers behind a load balancer), set a shared INTERNAL_TOKEN:

DJANGO_MCP = {
    "INTERNAL_TOKEN": "your-shared-secret-here",  # Use a strong random value
}

When INTERNAL_TOKEN is set, it is used instead of the auto-generated per-process token. All workers and the runmcp process must share the same value.

Tip: Generate a strong token with python -c "import secrets; print(secrets.token_urlsafe(32))".

2. Client-side: MCP client to the /mcp/ endpoint

The embedded /mcp/ view is a DRF APIView and inherits DEFAULT_AUTHENTICATION_CLASSES and DEFAULT_PERMISSION_CLASSES from REST_FRAMEWORK settings. If your DRF config requires authentication globally, MCP clients must also send auth headers:

{
  "mcpServers": {
    "my-app": {
      "type": "streamable-http",
      "url": "http://localhost:8000/mcp/",
      "headers": {
        "Authorization": "Token abc123..."
      }
    }
  }
}

Full authentication example

# settings.py
REST_FRAMEWORK = {
    "DEFAULT_AUTHENTICATION_CLASSES": [
        "rest_framework.authentication.TokenAuthentication",
    ],
    "DEFAULT_PERMISSION_CLASSES": [
        "rest_framework.permissions.IsAuthenticated",
    ],
}

DJANGO_MCP = {
    "NAME": "my-api",
    "HEADERS": {
        "Authorization": "Token abc123...",
    },
}
// .mcp.json (MCP client)
{
  "mcpServers": {
    "my-api": {
      "type": "streamable-http",
      "url": "http://localhost:8000/mcp/",
      "headers": {
        "Authorization": "Token abc123..."
      }
    }
  }
}

Endpoints

Path Description Controlled by
/mcp/ MCP protocol endpoint (JSON-RPC) MCP_PATH
/mcp/docs MCP Docs UI MCP_DOCS_ENABLED
/mcp/openapi.json MCP OpenAPI schema MCP_DOCS_ENABLED
/mcp/api/tools MCP tools JSON API (list) MCP_DOCS_ENABLED
/mcp/api/tools/{tool_name} MCP tool detail JSON API MCP_DOCS_ENABLED
/mcp/favicon MCP Docs favicon MCP_DOCS_ENABLED
/api/schema/ DRF OpenAPI schema (YAML) SWAGGER_ENABLED
/api/docs/ DRF Swagger UI SWAGGER_ENABLED

How It Works

  1. django_drf_mcp auto-injects drf-spectacular into INSTALLED_APPS and configures DEFAULT_SCHEMA_CLASS
  2. On startup, it generates an OpenAPI schema from all registered DRF views
  3. The base URL is injected into the OpenAPI schema servers list
  4. FastMCP converts each API endpoint into an MCP tool via OpenAPIProvider
  5. An httpx.AsyncClient is created with the configured BASE_URL and HEADERS for proxying tool calls
  6. Tools are served via STDIO, SSE, Streamable HTTP, or the embedded Django view
  7. When running under ASGI, fastmcp-docs provides interactive documentation at /mcp/docs

Management Command

python manage.py runmcp [options]
Option Default Description
--transport stdio Transport type: stdio, sse, streamable-http
--host 0.0.0.0 Bind host (for sse / streamable-http)
--port 8001 Bind port (for sse / streamable-http)
--base-url from settings Override BASE_URL for proxying API calls

Requirements

  • Python >= 3.10
  • Django >= 4.0
  • Django REST Framework

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured