django-stateless-mcp
Stateless MCP server for Django, enabling Django applications to interact via the Model Context Protocol.
README
django-stateless-mcp
A stateless Model Context Protocol server for Django, built on the 2026-07-28 spec. An MCP endpoint becomes an ordinary Django view — no sessions, no SSE, no sticky routing, and no dedicated single-process service.
- GitHub | PyPI | Documentation
- Created by Ben Atkinson | GitHub @BenA-SA
- MIT License
Why stateless
Before 2026-07-28, MCP needed a persistent connection: elicitation held an open stream and a blocked worker, so a reply that round-robined to another worker failed. The 2026-07-28 spec makes MCP plain request/response HTTP — Django's home turf — and turns elicitation into a retry any instance can serve. That is the change this package is built on; see Why stateless for the full story.
Usage
# myproject/mcp.py
from mcp.server.mcpserver import MCPServer
from django_stateless_mcp import request_state_security
server = MCPServer(
name="my-server",
version="1.0.0",
request_state_security=request_state_security(),
)
@server.tool()
def add(a: int, b: int) -> int:
"""Add two integers."""
return a + b
# urls.py
from django.urls import path
from django_stateless_mcp import mcp_view
from myproject.mcp import server
urlpatterns = [path("mcp/", mcp_view(server))]
That is the whole integration. The endpoint runs under both WSGI and ASGI.
What it provides
mcp_view(server)— serve anMCPServeras a stateless streamable-HTTP Django view, optionally requiring OAuth bearer auth via atoken_verifier.mcp.pyautodiscovery — adddjango_stateless_mcptoINSTALLED_APPSand each app'smcp.pyregisters tools, likeadmin.py.request_state_security()— key the SDK's elicitation-resume encryption fromSECRET_KEY, so elicitation survives a multi-worker deployment.django_request(ctx)— reach the authenticated Django request from inside a tool, with no global state.StructlogRequestLogger— optional flow-logging middleware.
Tool registration, elicitation, resources and prompts are the MCP SDK's own API; this package is the Django layer around it.
Supported versions
- Python 3.12–3.14
- Django 5.2 LTS and 6.0
mcp2.0.x
Each is exercised in CI, along with an advisory job tracking the SDK's git main.
Try it live
The repo ships a runnable example project. just demo-asgi starts it behind
four worker processes; the example README walks through
watching an elicitation started on one worker resume on another — the
package's thesis, observable with curl or any MCP client.
Documentation
Full documentation, including a worked elicitation example and the design decisions behind the package, is at https://django-stateless-mcp.readthedocs.io/.
Development
git clone git@github.com:Streamlined-Analytics/django-stateless-mcp.git
cd django-stateless-mcp
uv sync
uv run pytest # quick run: locked Django
uvx --with tox-uv tox run -f py313 # the full matrix: Django 5.2 + 6.0
just qa # format, lint, type check, test
just conformance # the official MCP conformance suite
Author
django-stateless-mcp was created in 2026 by Ben Atkinson.
Started from the audreyfeldroy/cookiecutter-pypackage template.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。