backend2mcp
Automatically converts Python web backends (FastAPI, Flask, Django) into MCP servers by exposing API routes as MCP tools with near-zero boilerplate.
README
backend2mcp
Convert any Python web backend into a fully functional MCP (Model Context Protocol) server automatically, with near-zero developer boilerplate.
What is this?
backend2mcp automatically exposes your existing Python web API routes as MCP tools. No HTTP servers, no proxies, no code generation—just import and run.
Installation
Install the core package:
pip install backend2mcp
Install with framework support:
pip install backend2mcp[fastapi] # FastAPI support
pip install backend2mcp[flask] # Flask support
pip install backend2mcp[django] # Django support
pip install backend2mcp[fastapi,flask] # Multiple frameworks
Quickstart
FastAPI
from fastapi import FastAPI
from backend2mcp.fastapi import MCPAdapter
app = FastAPI()
@app.get("/users/{id}")
async def get_user(id: int):
return {"id": id, "name": "Satyam"}
MCPAdapter(app).run()
Flask
from flask import Flask
from backend2mcp.flask import MCPAdapter
app = Flask(__name__)
@app.route("/users/<int:id>")
def get_user(id):
return {"id": id, "name": "Satyam"}
MCPAdapter(app).run()
Django
from django.urls import path
from backend2mcp.django import MCPAdapter
urlpatterns = [
path("users/<int:id>/", views.get_user),
]
MCPAdapter(urlpatterns=urlpatterns).run()
CLI Usage
# Auto-detect framework
backend2mcp run app:app
# Explicit framework
backend2mcp run app:app --framework fastapi
backend2mcp run app:app --framework flask
backend2mcp run app:app --framework django
Decorator Override
Customize tool behavior with @mcp_tool:
from backend2mcp.fastapi import MCPAdapter, mcp_tool
app = FastAPI()
@app.get("/users/{id}")
@mcp_tool(
name="get_user",
description="Get a user by their ID",
hidden=False
)
async def get_user(id: int):
return {"id": id}
Tool Naming
Tools are auto-named using a consistent convention:
| Route | Tool Name |
|---|---|
GET /users/{id} |
get_by_id |
POST /search |
post_search |
PUT /users/{id} |
put_by_id |
DELETE /users/{id} |
delete_by_id |
Authentication
backend2mcp provides flexible authentication support through pluggable AuthProvider classes.
No Authentication (Default)
Zero config, no auth required:
adapter = MCPAdapter(app) # Works without auth
Bearer Token Auth
from backend2mcp.core import BearerAuthProvider
auth = BearerAuthProvider(
token_header="Authorization", # Header name
token_prefix="Bearer", # Token prefix
validate_tokens=["secret1", "secr2"] # Optional whitelist
)
adapter = MCPAdapter(app, auth_provider=auth)
API Key Auth
from backend2mcp.core import APIKeyAuthProvider
auth = APIKeyAuthProvider(
header_name="X-API-Key", # Header name
query_param="api_key", # Query param name
valid_keys=["key1", "key2"] # Optional whitelist
)
adapter = MCPAdapter(app, auth_provider=auth)
Custom Headers Injection
from backend2mcp.core import HeaderInjectionAuthProvider
auth = HeaderInjectionAuthProvider(
static_headers={
"X-Custom-Header": "value",
"Authorization": "Bearer static-token"
}
)
adapter = MCPAdapter(app, auth_provider=auth)
Combining Providers
from backend2mcp.core import (
BearerAuthProvider,
APIKeyAuthProvider,
combine_providers
)
auth = combine_providers(
BearerAuthProvider(),
APIKeyAuthProvider()
)
adapter = MCPAdapter(app, auth_provider=auth)
Accessing Auth in Handlers
Auth context is injected into handlers:
from backend2mcp.fastapi import MCPAdapter
from backend2mcp.core import BearerAuthProvider, AuthContext
app = FastAPI()
auth = BearerAuthProvider()
@app.get("/users/{id}")
async def get_user(id: int, auth_context: AuthContext = None):
headers = auth_context.headers if auth_context else {}
user = get_user_from_db(id, headers=headers)
return user
adapter = MCPAdapter(app, auth_provider=auth)
Architecture
backend2mcp/
├── core/ # Shared implementation
│ ├── adapter.py # BaseAdapter abstract interface
│ ├── auth.py # Auth providers (Bearer, API Key, etc.)
│ ├── server.py # MCP server implementation
│ ├── schema.py # Schema conversion utilities
│ └── exceptions.py # Structured exceptions
├── fastapi/ # FastAPI adapter
├── flask/ # Flask adapter
├── django/ # Django adapter
└── cli/ # Typer CLI
Core Abstractions
BaseAdapter: Abstract interface all framework adapters implementAuthProvider/AuthContext: Pluggable authentication systemMCPServer: Handles MCP protocol using officialmcpSDK- Tool Execution: Direct handler invocation (no HTTP calls)
- Schema Generation: Pydantic-integrated JSON Schema extraction
Writing a New Adapter
To add support for another framework:
- Create
backend2mcp/framework/ - Implement
BaseAdapterinterface:get_routes()- Extract all routes from the frameworkintrospect_route()- Convert a route toToolInfoexecute_tool()- Call handler with resolved argumentsbuild_tool_name()- Generate MCP-safe tool names
- Export
MCPAdapterfrom the subpackage
from backend2mcp.core.adapter import BaseAdapter, ToolInfo
class MCPAdapter(BaseAdapter):
def get_routes(self) -> list[tuple]:
# Your route extraction logic
pass
def introspect_route(self, path, method, handler, config) -> ToolInfo:
# Your schema extraction logic
pass
def execute_tool(self, handler, arguments, context=None):
# Your handler invocation logic
pass
def get_app(self):
# Return underlying framework object
pass
def build_tool_name(self, http_method, path):
# Your naming convention
pass
Roadmap
| Phase | Frameworks |
|---|---|
| Phase 1 (this release) | FastAPI, Flask, Django |
| Phase 2 | Express, NestJS, Fastify |
| Phase 3 | Spring Boot, Quarkus |
| Phase 4 | Gin, Fiber, Echo |
License
MIT
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。