ApiBridge MCP
A bridge that converts HTTP APIs into MCP tools with dynamic toolkit loading, flexible authentication, and automatic documentation.
README
📘 ApiBridge MCP
基于 Python 3.10+ 和 FastAPI 开发的HTTP API桥接MCP服务,支持多工具集(Toolkits)、动态工具加载、多种认证方式和自动API文档生成。
🚀 功能特性
- 多工具集支持: 从
toolkits/目录动态加载多个工具集(每个 JSON 文件一个工具集),每个工具集有独立的配置和工具。 - 动态工具加载: 根据JSON配置文件自动创建和注册工具,无需修改代码。
- 灵活的认证机制: 支持
http basic和oauth2_jwt两种认证方式,可为每个工具集独立配置。 - 中间件驱动的路由: 通过
ToolkitRoutingMiddleware中间件,根据Toolkit-CodeHTTP头动态路由到正确的工具集。 - 内置工具: 提供
current_time,get_api_docs,api_version等实用内置工具。 - 自动文档生成: 支持通过命令行生成所有或单个工具集的Markdown格式API文档。
- 异步HTTP客户端: 使用
aiohttp进行高效的异步外部API调用。 - 容器化部署: 提供完整的Docker和Docker Compose配置,方便快速部署。
🛠️ 环境要求
- Python 3.10+
- pip
- pnpm (或 npm/yarn)
- Docker(可选)
项目结构
apibridge-mcp-service/
├── .env.example # 环境变量示例文件
├── README.md # 项目文档
├── app/ # 主应用目录
│ ├── auth/ # 认证模块 (http_basic, oauth2_jwt)
│ ├── config/ # 配置模块 (settings.py)
│ ├── mcp_server.py # FastAPI应用和MCP服务器核心
│ ├── tools/ # 内置工具实现
│ └── utils/ # 实用工具 (doc_generator, tool_loader, middlewares)
├── toolkits/ # 工具集配置目录(每个文件代表一个工具集)
│ └── *.json
├── docs/ # 生成的文档存放目录
├── tests/ # 测试代码目录
├── main.py # 应用入口
└── requirements.txt # Python依赖
⚙️ 安装与配置
-
克隆项目
-
安装依赖
pip install -r requirements.txt -
配置工具集
在 toolkits/ 目录下创建JSON文件来定义你的工具集。例如, my_toolkit.json:
```json
{
"code": "my_toolkit",
"name": "My Custom Toolkit",
"description": "A toolkit for custom operations.",
"auth": {
"type": "http_basic",
"user_name": "myuser",
"password": "mypass"
},
"tools": [
{
"name": "my_tool",
"description": "Does something awesome.",
"path": "/custom/endpoint",
"method": "POST",
"parameters": { ... },
"response": { ... }
}
]
}
```
-
配置环境变量 (可选)
创建
.env文件并根据.env.example的内容进行配置,可以覆盖settings.py中的默认值。
运行服务
python main.py
服务默认启动在 http://0.0.0.0:8200。
🚀 使用方法
工具调用
所有工具调用都通过 /tools/ 端点进行。你需要提供 Toolkit-Code HTTP头来指定要使用的工具集。
请求示例:
POST /tools/ HTTP/1.1
Host: localhost:8200
Content-Type: application/json
Toolkit-Code: my_toolkit
{
"tool_name": "my_tool",
"parameters": { ... }
}
内置工具
current_time: 获取当前时间。get_api_docs: 获取API文档。可使用toolkit_code参数获取特定工具集的文档。api_version: 获取服务版本。
📚 API文档生成
通过命令行生成Markdown格式的API文档。
-
生成所有工具集的文档:
python main.py --doc -
生成特定工具集的文档:
python main.py --doc --toolkit your_toolkit_code
文档将生成在 docs/ 目录下。
✅ 测试
项目包含一套完整的单元测试和集成测试。运行测试:
pytest
🐳 Docker部署
使用提供的脚本构建和运行Docker容器。
-
构建镜像:
cd docker ./build_image.sh -
使用Docker Compose运行:
cd docker docker-compose up -d
📄 日志
服务日志提供了详细的请求、响应和错误信息,方便调试和监控。
🧰 工具列表(简版) / Tools List (Simple)
- 路由:GET
/debug/tools-simple - 作用:返回简化的工具映射,包含
toolkitCode、toolCode、internalName以及description - 查询参数:
toolkit(可选):仅列出指定工具包的工具tool(可选):仅列出指定工具名的工具
- 冲突提示:当同名
toolCode出现在多个工具包中,会在返回中附带conflicts列表(字段:toolCode、toolkits),便于客户端避免歧义。
查询参数 / Query params:
- toolkit(string,可选):按工具包过滤
- tool(string,可选):按工具名过滤
- page(number,可选,默认 1):分页页码
- pageSize(number,可选,默认 50,最大 500):分页大小
- sortBy(string,可选,默认 toolCode):可选 toolCode | toolkitCode | internalName
- sortOrder(string,可选,默认 asc):可选 asc | desc
- unique(boolean,可选):仅返回不冲突(唯一)的工具名
- conflictsOnly(boolean,可选):仅返回存在跨工具包同名冲突的工具条目
示例(PowerShell):
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple" | ConvertTo-Json -Depth 3
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple?toolkit=sale-solution-tools" | ConvertTo-Json -Depth 3
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple?tool=current_time" | ConvertTo-Json -Depth 3
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple?page=1&pageSize=20&sortBy=toolkitCode&sortOrder=desc" | ConvertTo-Json -Depth 3
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple?unique=true" | ConvertTo-Json -Depth 3
Invoke-RestMethod "http://127.0.0.1:8201/debug/tools-simple?conflictsOnly=true" | ConvertTo-Json -Depth 3
English:
- Endpoint: GET
/debug/tools-simple - Purpose: Return a simplified mapping with
toolkitCode,toolCode,internalName, anddescription - Query params:
toolkit(optional),tool(optional)page(optional, default 1),pageSize(optional, default 50, max 500)sortBy(optional, default toolCode): toolCode | toolkitCode | internalNamesortOrder(optional, default asc): asc | descunique(optional): only unique toolCodesconflictsOnly(optional): only conflicting toolCodes
- Conflicts: If the same
toolCodeappears across multiple toolkits, aconflictsarray is included.
📦 工具包列表(简版) / Toolkits List (Simple)
- 路由:GET
/debug/toolkits-simple - 返回:每个工具包的
toolkitCode、工具数量toolsCount和内置工具数量builtinCount
示例(PowerShell):
Invoke-RestMethod "http://127.0.0.1:8201/debug/toolkits-simple" | ConvertTo-Json -Depth 3
English:
- Endpoint: GET
/debug/toolkits-simple - Returns:
toolkitCode,toolsCount,builtinCountfor each toolkit
🔗 MCP Stream 调用约定 / MCP Stream Protocol
- 服务端内部注册名统一为:
toolkitCode__toolCode(避免跨工具包重名冲突,便于统计与治理) mcp:list-tools响应返回组合名toolkitCode__toolCode列表(与内部注册保持一致,避免客户端拿到纯名后底层出现 Unknown tool)mcp:call-tool支持两种调用方式:- 组合名:传
tool_name=toolkitCode__toolCode(推荐;同时提供Toolkit-Code以校验工具集存在与权限) - 纯名:传
tool_name=toolCode(兼容;中间件会根据Toolkit-Code自动改写为组合名)
- 组合名:传
- Header 要求:
Accept必须包含application/json, text/event-stream,Content-Type使用application/json - 会话要求:若返回 “Missing session ID”,请先进行一次会话初始化(例如通过浏览器打开
/dashboard或按你的会话流程),再调用mcp:call-tool
示例(PowerShell):
# 初始化(内置方法)
Invoke-WebRequest -Uri "http://127.0.0.1:8200/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123"; "toolkit-code"="sale-solution-tools"; "Accept"="application/json, text/event-stream" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"init-1","method":"initialize","params":{"clientName":"CherryStudio","clientVersion":"1.0"}}'
# 列表工具(内置方法)
Invoke-WebRequest -Uri "http://127.0.0.1:8200/mcp/stream?toolkit-code=sale-solution-tools" -Method POST -Headers @{ "x-client-key"="abc123"; "Accept"="application/json, text/event-stream" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"list-1","method":"mcp:list-tools","params":{}}'
# 调用具体工具(纯 toolCode + Toolkit-Code)
Invoke-WebRequest -Uri "http://127.0.0.1:8200/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123"; "toolkit-code"="sale-solution-tools"; "Accept"="application/json, text/event-stream" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"call-1","method":"mcp:call-tool","params":{"tool_name":"current_time","arguments":{}}}'
English:
- Server internal registration uses
toolkitCode__toolCodeto avoid cross-toolkit naming conflicts and keep stats consistent. mcp:list-toolsreturns combined namestoolkitCode__toolCode(consistent with server registration; avoids Unknown tool when calling).mcp:call-toolsupports both:- Combined name: send
tool_name=toolkitCode__toolCode(recommended; still provideToolkit-Codefor toolkit validation) - Pure name: send
tool_name=toolCode(compatible; middleware rewrites usingToolkit-Code)
- Combined name: send
- Headers:
Acceptmust includeapplication/json, text/event-stream,Content-Typeshould beapplication/json. - Session: If you receive “Missing session ID”, initialize a session first (e.g., open
/dashboardin browser or follow your session init flow), then callmcp:call-tool.
🔐 日志脱敏策略 / Log Sanitization Policy
- 为满足安全规范,服务会对以下 Header/字段进行脱敏输出:
authorization、x-client-key、x-api-key、x-token、token、password - 日志中的敏感值统一显示为
***
English:
- For security, logs mask sensitive headers/fields:
authorization,x-client-key,x-api-key,x-token,token,password. - Masked values appear as
***in logs.
❗ 错误码与常见排错 / Error Codes & Troubleshooting
说明 / Notes:以下错误示例为典型形态,实际返回格式可能因 FastAPI 的异常处理而有所差异(可能返回 { "detail": "..." });建议客户端同时兼容 detail 和自定义 error.message 两种字段。
常见场景 / Common Cases:
- 缺失 Toolkit-Code(工具调用)/ Missing Toolkit-Code (tool call)
- 条件:调用
mcp:call-tool时未在 Header 或 Query 提供Toolkit-Code/toolkit-code - 请求示例(PowerShell):
Invoke-WebRequest -Uri "http://127.0.0.1:8201/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"call-1","method":"mcp:call-tool","params":{"tool_name":"current_time","arguments":{}}}'
- 预期响应(400):
{
"error": {
"code": 400,
"message": "Toolkit-Code required for tool call (send header 'Toolkit-Code' or query 'toolkit-code')",
"hint": "Use pure toolCode plus Toolkit-Code. Example: header Toolkit-Code=sale-solution-tools"
}
}
或(FastAPI 默认)
{ "detail": "Toolkit-Code is required when calling tools" }
- 工具名建议 / Tool naming recommendation
- 推荐:使用组合名
toolkitCode__toolCode;纯名也兼容(中间件会根据Toolkit-Code自动改写)。 - 请始终提供
Toolkit-Code(header 或 query),以便服务在鉴权与治理上进行工具集校验。
- 工具包不存在 / Toolkit JSON not found
- 条件:
Toolkit-Code指向不存在的工具包 JSON - 请求示例:
Invoke-WebRequest -Uri "http://127.0.0.1:8201/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123"; "toolkit-code"="unknown-toolkit" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"call-3","method":"mcp:call-tool","params":{"tool_name":"current_time","arguments":{}}}'
- 预期响应(404):
{
"error": {
"code": 404,
"message": "Toolkit not found: unknown-toolkit",
"hint": "Check available toolkits via /debug/toolkits-simple or /debug/tools-simple"
}
}
或(FastAPI 默认)
{ "detail": "Toolkit not found: unknown-toolkit" }
- 工具不存在(在指定工具包下)/ Tool not found in toolkit
- 条件:
tool_name在该Toolkit-Code下未注册 - 请求示例:
Invoke-WebRequest -Uri "http://127.0.0.1:8201/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123"; "toolkit-code"="sale-solution-tools" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"call-4","method":"mcp:call-tool","params":{"tool_name":"non_exist_tool","arguments":{}}}'
- 预期响应(404):
{
"error": {
"code": 404,
"message": "Tool not found in toolkit",
"hint": "Use /debug/tools-simple?toolkit=sale-solution-tools to confirm available toolCode"
}
}
或(FastAPI 默认)
{ "detail": "Tool not found in toolkit" }
- 内置 JSON-RPC 方法(不需要 tool_name)/ Built-in methods (no tool_name required)
- 条件:
initialize、mcp:list-tools等方法无需tool_name,会直接透传到 FastMCP - 请求示例:
Invoke-WebRequest -Uri "http://127.0.0.1:8201/mcp/stream" -Method POST -Headers @{ "x-client-key"="abc123"; "toolkit-code"="sale-solution-tools" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"init-1","method":"initialize","params":{"clientName":"CherryStudio","clientVersion":"1.0"}}'
Invoke-WebRequest -Uri "http://127.0.0.1:8201/mcp/stream?toolkit-code=sale-solution-tools" -Method POST -Headers @{ "x-client-key"="abc123" } -ContentType "application/json" -Body '{"jsonrpc":"2.0","id":"list-1","method":"mcp:list-tools","params":{}}'
网络错误 / Network Errors:
- “无法连接到远程服务器”:检查服务是否运行、端口是否开放(8201)、本机防火墙策略、代理软件拦截;可使用
netstat -ano | findstr 8201验证监听状态。
English:
Notes: Response formats may vary ({"detail":"..."} vs. custom error.message). Clients should handle both.
Common cases:
- Missing Toolkit-Code (tool call)
- Condition: Calling
mcp:call-toolwithoutToolkit-Code(header or query) - Expected 400:
{ "error": { "code": 400, "message": "Toolkit-Code required", "hint": "Send header Toolkit-Code or query toolkit-code" } }
- Tool naming recommendation
- Preferred: use combined names
toolkitCode__toolCode; pure names are compatible (middleware rewrites usingToolkit-Code). - Always provide
Toolkit-Code(header or query) so the service can validate toolkit in auth and governance.
- Toolkit JSON not found
- Condition:
Toolkit-Codepoints to a non-existent toolkit - Expected 404:
{ "error": { "code": 404, "message": "Toolkit not found: unknown-toolkit" } }
- Tool not found in toolkit
- Condition:
tool_namenot registered under the given toolkit - Expected 404:
{ "error": { "code": 404, "message": "Tool not found in toolkit" } }
- Built-in JSON-RPC methods
initializeandmcp:list-toolsrequire notool_name; they are passed through to FastMCP.
Network errors:
- "Cannot connect to remote server": ensure the service is running, port 8201 is open, firewall allows local connections, and no proxy blocks requests; use
netstat -ano | findstr 8201to verify the listening state.
Recommended Servers
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.