mcp-pipeline-agent

# 🚀 MCP Pipeline Agent > An autonomous software engineering development and testing toolchain based on the Model Context Protocol (MCP) and Cline Agent. [](https://opensource.org/licenses/MIT) [](https://www.python.org/) [](https://modelcontextprotocol.io/) `mcp-pipeline-agent` is a system-level autonomous development Agent toolchain powered by the **Model Context Protocol (MCP)** and driven by explicit state machines. It enables AI to automatically handle project scaffolding, code quality scanning, automated bug fixing, web-based closed-loop testing, and pipeline verification following standardized software engineering practices. --- ## 🌟 Key Features - **Full Domain Coverage**: Supports full-lifecycle engineering across Frontend (Vue 3 + uni-app), Backend (ThinkPHP), and Admin Dashboard (FastAdmin) with 16 core tools integrated. - **Progressive Tool Discovery**: Employs meta-tool dynamic retrieval and role-based filtering, avoiding context window bloat and cutting initial token consumption by ~**85%**. - **7-Stage Lifecycle State Machine**: State control governed explicitly via `pipeline_tasks`, keeping long-horizon development tasks strictly deterministic and manageable. - **Dual-Engine Web Testing Loop**: - **First Line of Defense**: Playwright API silent listener engine for millisecond-level interception of Console errors, HTTP 4xx/500 network anomalies, and DOM assertion failures. - **Fallback Safeguard**: Computer Use vision-control engine invoking VLMs for screen coordinate prediction and UI interaction when handling Canvas element renderers or non-DOM components. - **Self-Healing Loop**: Automatically generates structured **Diagnostic Bundles** upon capturing test failures, triggering targeted hot-reloads, auto-repairs, and regression testing retries. --- ## 🛠️ System Architecture ```text [Requirement Extraction] ➔ [Requirement Analysis] ➔ [Tool/Skill Loading] ➔ [Code Construction] │ [Delivery Completed] ◄─ [Retry Limit / Human Intervention] ◄─ [Self-Healing] ◄─ [Dual-Engine Web Test]