Make Automation Toolkit

制作自动化工具包

专业的Python SDK和参考指南，用于部署、管理和安排Make.com自动化，包括REST API、MCP服务器和AI代理。

   

______________________________________________________________________

演示

https://github.com/cognizonline/make-automation-toolkit/assets/make_skills.mp4

______________________________________________________________________

这是什么

Make.com是一个无代码自动化平台。该工具包充分展现了其强大功能 作为 代码 --让工程师：

• 从版本控制的JSON蓝图部署生产场景
• 通过类型化的Python客户端管理数据存储、Webhook和连接
• 使用LLM提供者和工具批准策略配置AI代理
• 将场景作为可调用工具公开给任何兼容MCP的AI助手

______________________________________________________________________

特性

______________________________________________________________________

存储库结构

make-automation-toolkit/
├── src/
│   ├── make_client.py              # MakeClient + MakeDeployer SDK
│   ├── validate_blueprint.py       # blueprint schema validator
│   ├── blueprints/
│   │   ├── schema.json             # JSON Schema for blueprint validation
│   │   ├── ai_local_agent.json     # scenario-embedded agent template
│   │   ├── basic_webhook.json
│   │   ├── ecommerce_order_processing.json
│   │   └── customer_tracking.json
│   └── examples/
│       ├── 01_deploy_scenario.py
│       ├── 02_manage_data_store.py
│       ├── 03_configure_agent.py
│       ├── 04_setup_mcp.py
│       ├── 05_full_deployment.py
│       ├── 06_deploy_scenario_agent.py   # scenario-embedded AI agent
│       ├── 07_builtin_ai_tools.py        # agent using Make built-in AI modules
│       └── 08_mcp_toolbox_workflow.py    # MCP Toolbox governed tool pattern
├── skill/
│   ├── README.md
│   └── make-automation-skill.md    # single-file AI context reference
├── prompts/
│   ├── README.md
│   ├── lead_generation.md
│   ├── customer_support.md
│   ├── document_processing.md
│   ├── research_summarisation.md
│   ├── data_enrichment.md
│   ├── document_and_media_processing.md  # make-ai-extractors patterns
│   └── _template.md
├── docs/
│   ├── quickstart.md
│   ├── authentication.md
│   ├── mcp-integration.md
│   ├── mcp-toolboxes.md            # governed MCP access for production
│   ├── ai-agents.md
│   └── best-practices.md
├── tests/
│   └── test_make_client.py
├── assets/
│   └── make_skills.mp4
├── .github/workflows/ci.yml
├── pyproject.toml
├── requirements.txt
└── README.md

______________________________________________________________________

快速开始

# 1. Clone
git clone https://github.com/cognizonline/make-automation-toolkit.git
cd make-automation-toolkit
pip install -r requirements.txt

# 2. Set credentials
export MAKE_API_TOKEN="your-token"
export MAKE_ZONE="eu1.make.com"
export MAKE_TEAM_ID="123"

# 3. Run any example
python src/examples/01_deploy_scenario.py

看 docs/quickstart.md 进行完整设置。

______________________________________________________________________

SDK使用

客户

from src.make_client import MakeClient

client = MakeClient(
    api_token="your-token",
    zone="eu1.make.com",
    team_id=123,
)

# Scenarios
scenarios  = client.list_scenarios()
scenario_id = client.create_scenario(blueprint, scheduling={"type": "on-demand"})
client.activate_scenario(scenario_id)
result = client.run_scenario(scenario_id, data={"key": "value"})

# Data stores
structure_id = client.create_data_structure("Orders", spec)
store_id     = client.create_data_store("Order Store", structure_id)
client.add_record(store_id, {"order_id": "ORD-001", "status": "pending"})
records = client.list_records(store_id)

# Webhooks
hook = client.create_hook("Inbound Events")
print(hook["url"])  # send events here

# AI Agents
agent_id = client.create_agent(config)
reply    = client.run_agent(agent_id, messages=[{"role": "user", "content": "..."}])

高级部署人员

from src.make_client import MakeClient, MakeDeployer

deployer = MakeDeployer(client)

# Deploy a scenario as an MCP tool in one call
scenario_id = deployer.deploy_mcp_tool(blueprint, inputs, outputs, activate=True)

# Deploy a scenario + data store together
result = deployer.deploy_with_datastore(blueprint, "My Store", structure_spec)

______________________________________________________________________

蓝图

蓝图是下面的纯JSON文件 src/blueprints/每个蓝图 描述了完整的场景流，包括模块、映射器、过滤器和 错误处理程序。

{
  "name": "Basic Webhook Trigger",
  "flow": [
    { "id": 1, "module": "gateway:CustomWebhook", ... },
    { "id": 2, "module": "http:ActionSendData",   ... }
  ],
  "metadata": {
    "version": 1,
    "scenario": {
      "roundtrips": 1,
      "maxErrors": 3,
      "autoCommit": true,
      ...
    }
  }
}

每次推送时，蓝图都会在CI中得到验证。

______________________________________________________________________

MCP集成

将任何Make场景作为可由Claude、Cursor或任何MCP调用的工具公开 客户。三个要求：按需调度、主动、类型化I/O。

scenario_id = deployer.deploy_mcp_tool(blueprint, inputs, outputs)

然后添加到您的Claude配置中：

{
  "mcpServers": {
    "make": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://eu1.make.com/mcp/u/YOUR_MCP_TOKEN/sse"]
    }
  }
}

看 docs/mcp-integration.md 完整的指南。

______________________________________________________________________

文档

______________________________________________________________________

错误处理和重试

所有SDK方法都包括内置的指数回退 429 Too Many Requests 和瞬态 5xx 错误：

# Configurable per-call
result = client._request("GET", "/scenarios", max_retries=5)

常见状态代码及其含义： docs/best-practices.md.

______________________________________________________________________

许可证

麻省理工学院——见 许可证.

______________________________________________________________________

致谢

建在 Make.com REST API v2 和那个 制作MCP服务器.