技术参考文档

PETP — 流水线 · 执行 · 任务 · 处理器

基于 Python 的可配置任务运行器、执行引擎和 MCP 工具服务器。PET = Pipeline-Execution-Task，层级模型。末尾的 P = Processor，与每个任务一一对应。

Pipeline 1:n Execution Execution 1:n Task Task 1:1 Processor

运行模式

🖥️

桌面 GUI

基于 wxPython 的完整可视化运行时。支持交互式 RPA、本地开发和手动触发执行。

python PETP.py

⚙️

后台 / JOB模式

无 GUI，自动检测后台 Chrome，适用于定时任务、CLI 流水线和服务器自动化。

python PETP_backgroud.py

🐳

Docker

多架构镜像（arm64 构建 → amd64 运行），无头模式，支持 CI/CD 和 NAS 部署。

docker run -p 8866:8866 petp

🔌

MCP 工具服务器

将执行任务暴露为带类型的 MCP 工具，同时支持 stdio 和 Streamable-HTTP 传输方式。

GET /mcp · port 8866

处理器库

分类	能力
Browser Automation Selenium	Navigate, go back, fullscreen, close Chrome. Find element(s) then click / key-in / collect. Batch find with skip. iFrame, cookies, screenshot. Convert Selenium IDE recordings to PETP tasks.
SSH / SFTP Paramiko	Create SSH / SFTP sessions. Run remote commands. Upload / download files.
File & Folder	Open, write, delete, read text. Find files / latest file. Watch & auto-move. ZIP / UNZIP. File-chooser dialog.
Data & Spreadsheet	Read CSV / Excel. Write to Excel. CSV to XLSX. Collect, filter, group-by, mapping, masking, conversion. Merge collections.
Database CRUD	MySQL, PostgreSQL, SAP HANA, SQLite — unified `DB_ACCESS` processor.
AI / LLM	DeepSeek, Google Gemini, Ollama (local), Zhipu Z.AI — each with Setup + Q&A + MCP-tool calling processors.
MCP	Expose PETP as standard MCP Tool Server (Streamable-HTTP). MCP client processors for all LLM providers.
HTTP / Network	Configurable HTTP requests. Extract response keys. Built-in HTTP service (port 8866). OAuth2 / PKCE.
String Utilities	Encode / decode (Base64, URL…). Hash (MD5, SHA256…).
Mouse & GUI PyAutoGUI	Click, scroll, query position at absolute or relative coordinates.
Email	Send email via SMTP with attachments.
Data Visualization	Charts and plots via Matplotlib.
Execution Control	Init / check params. Nested execution. Conditional stop. Wait / sleep. Reload log config. Read JSON. Run shell commands. Run JavaScript.

HTTP 服务与 MCP 端点

内置 HTTP 服务器 — 端口 8866

方法	端点	说明
GET	/	PETP HTTP service home
GET	/mcp	MCP Tool Server — Streamable-HTTP transport
GET	/health	Health check (Docker / container)
GET	/petp/tools	List all exposed MCP tools
POST	/petp/exec	Trigger execution or pipeline (async)
GET	/petp/result	Poll async result by `?request_id=<id>`
POST	/mcp	MCP Tool Server — direct POST endpoint

MCP Inspector 配置

将任何 MCP Inspector 或兼容 MCP 的客户端直接连接到 PETP 的 HTTP 服务器。

传输类型：Streamable HTTP
URL：http://localhost:8866/mcp
本地开发无需身份验证
Claude Code、Cursor、VS Code Copilot——在设置中添加为 MCP 服务器

# Claude Code — add PETP as MCP server
# .claude/settings.json or mcpServers config

"mcpServers": {
  "petp": {
    "type": "http",
    "url": "http://localhost:8866/mcp"
  }
}

# Or trigger via HTTP POST (no MCP client needed)
POST http://localhost:8866/petp/exec

{
  "wait_for_result": "true",
  "action": "execution",
  "params": {
    "execution": "YOUR_EXECUTION_NAME",
    "fromHTTPService": "true"
  }
}

依赖分组

Modular requirements/ structure — install only what you need. Uses uv (10-100× faster than pip) or standard pip.

core.txt

pyyaml, croniter, cryptocode

web-automation.txt

selenium, urllib3, Pillow

http-client.txt

requests, httpx, httpx-sse

ssh-sftp.txt

paramiko

excel-data.txt

openpyxl, pandas

database.txt

psycopg, mysql-connector, hdbcli

ai-deepseek.txt

openai (DeepSeek compat.)

ai-gemini.txt

google-genai

ai-ollama.txt

ollama (local LLM)

ai-zhipu.txt

zai (Zhipu Z.AI)

mcp.txt

mcp protocol

http-service.txt

fastapi, uvicorn

javascript.txt

pythonmonkey

gui-wxpython.txt

wxpython 4.3.x

gui-automation.txt

pyautogui, pyperclip

chart.txt

matplotlib

安装选项

Full (GUI desktop): requirements.txt
Background service: requirements-nogui.txt
Docker / headless: requirements-docker.txt
Custom: combine any group files with -r

# Recommended: uv (fast)
pip install -U uv
uv pip install -r requirements.txt

# Custom combination
uv pip install \
  -r requirements/core.txt \
  -r requirements/mcp.txt \
  -r requirements/http-service.txt

# Pin versions for reproducible builds
uv pip compile requirements.txt \
  -o requirements.lock

项目结构

目录 / 文件	说明
PETP.py	GUI desktop entry point
PETP_backgroud.py	Headless / background entry point
config/petpconfig.yaml	Main runtime configuration
core/executions/	YAML execution definitions
core/processors/	Processor implementations — one .py per task type
core/pipelines/	YAML pipeline definitions
core/runtime/	Background / no-GUI runtime logic
core/definition/	YAML reader, Selenium IDE converter
httpservice/	HTTP server, MCP handler, request routing
mvp/	GUI layer (Model-View-Presenter with wxPython)
utils/	Utilities — Selenium, Excel, Date, OS, Logger, Paramiko
webapp/	Flask web application (this site)
webdriver/	Platform ChromeDriver binaries