Releases: keta1930/agent-graph
Release list
mcp-agent-graph 3.0
1️⃣ 系统级 Agent
系统级Agent将帮助用户定制Agent Workflow/Agent Graph与 MCP 工具
2️⃣ 可视化图编辑器
前端创作智能体工作流，所见即所得
3️⃣ 图嵌套图
Agent复用性，任何图都可作为其他图的节点使用，构建分层智能系统
4️⃣ 任务调度系统
支持定时、周期性执行Agent，批量并发处理
5️⃣ 图转 MCP 服务
一键导出智能体为标准 MCP 服务，可被 Claude、Cline 等调用
6️⃣ 智能体交易与传递
完整的智能体打包、分享和部署解决方案
7️⃣ Python SDK 深度集成
pip install mcp-agent-graph 使用 Python 构建Agent。
8️⃣ 提示词注册管理
一站式提示词管理，注册复用提示词模板
V2.0 new release intro
1️⃣ From need to Agent
This is an amazing feature! AI-generated agents. Previously, you might need to write extensive code to design agents, or orchestrate workflows in the frontend by adding nodes one by one. mcp-agent-graph provides a solution: simply write down your requirements, and the system will generate an excellent graph for you!
From need to graph, it might take only 3 minutes, or even less! Once the graph is generated, it will appear on your canvas (agent). You can view each node in the graph, the overall process, and the readme file! Click on each node to see which tools (mcp server) it selected, prompts, models, context passing between nodes... If it doesn't meet your requirements, you can use the AI graph optimization feature to tell the AI your updated requirements, and it will adjust the process, add or remove nodes, modify prompts or tool calls for you.
2️⃣ AI-Generated MCP Tools (From need to MCP)
This is a future-oriented feature! Create custom MCP tools through natural language descriptions.
Traditional MCP tool development requires extensive programming knowledge and understanding of the MCP protocol. mcp-agent-graph breaks this barrier: simply describe what you want the tool to do, and AI will generate a complete, production-ready MCP tool for you!
From description to deployment in minutes! Just provide requirements, and the system will: automatically generate Python code following MCP standards, create proper virtual environments with all dependencies, handle port management and conflict detection, automatically register the tool to your system configuration, provide comprehensive documentation and usage examples. The generated tools are immediately ready for use in your agents or can be shared with the community. This democratizes MCP tool development, making it accessible to everyone regardless of technical background.
3️⃣ Nested Graphs (Hierarchical World)
This is an architectural innovation! Building "Agent within Agent" hierarchical intelligent systems. Traditional workflows are often flat, and when systems become complex, they become difficult to manage and maintain. mcp-agent-graph introduces the concept of nested graphs: any complete graph can be used as a single node in another graph! This creates infinite possibilities.
Hierarchical design, unlimited scalability! You can first build a "document analysis" graph containing document parsing, content extraction, format conversion nodes. Then encapsulate this entire graph as a single node for use in a larger "knowledge management" graph. This hierarchical design allows you to: build reusable agent modules, manage complex large-scale systems, achieve true modular development. Each layer has clear responsibility boundaries, maintaining system integrity while having extremely strong maintainability.
4️⃣ Graph to MCP Server
This is a standardization feature! Export agent graphs as standard MCP services. In the AI tool ecosystem, interoperability between different platforms and frameworks has always been a challenge. mcp-agent-graph provides graph-to-mcp functionality: one-click export of any graph as a standard MCP server Python script!
Build once, run everywhere! The exported MCP server fully complies with MCP protocol standards and can be directly called by Claude Desktop, Cline, cursor, and other AI applications or any MCP-supporting systems. Your agent instantly becomes a widely integrable tool. The exported script includes complete dependency management, configuration files, and installation instructions, allowing recipients to deploy immediately. This lays the foundation for agent standardization and ecosystem development.
5️⃣ Visual Graph Editor
Canvas as Code! You can build complex agent workflows simply by dragging nodes and connecting lines on the visual canvas. What you see is what you get, design is development! Each node has rich configuration options, allowing you to set prompts, select models, configure tool calls, and define input-output relationships directly in the interface. The connections between nodes clearly show data flow and execution order, making complex logic clear at a glance. Real-time preview functionality lets you view the execution effects of your current design at any time.
6️⃣ Node as Agent
Every node is an independent agent. Each node in the graph has complete Agent capabilities! Every node can call tools and handle complex tasks. Microservice-oriented agent architecture, each node is an expert! You can configure specialized role prompts for each node, making it an expert in specific domains. One node can be a data analyst, another can be a content creator, and a third can be a decision maker. They gain powerful tool capabilities through MCP servers, such as accessing file systems, web searching, performing calculations, etc. Nodes collaborate through context passing, forming a powerful agent team.
7️⃣ Agent Trading and Transfer
This is an ecosystem feature! Complete agent packaging, sharing, and deployment solution. In the current AI development environment, sharing a complete agent system often requires complex environment configuration, dependency installation, and documentation, greatly limiting agent propagation and reuse. mcp-agent-graph provides complete agent lifecycle management: packaging agent systems and all their dependencies into self-contained, portable units.
One-click packaging, one-click deployment, agent ecosystem! The system automatically generates comprehensive README documentation, detailing agent functionality, requirements, and usage methods. Recipients don't need to understand complex technical details to quickly understand and deploy your agents. This feature provides a complete solution for agent marketplace trading, team collaboration, and open-source sharing. You can easily: share professional tools with colleagues, deliver custom solutions to clients, contribute your creations to the open-source community.
8️⃣ Python SDK Deep Integration
This is a dual-wheel development mode! Perfect combination of frontend visual design and backend code execution. mcp-agent-graph provides through Python SDK: frontend drag-and-drop design, backend code execution! Perfect fusion of design and development, both visual and code-controllable!
You can quickly design and debug agent graphs in the frontend visual interface, then install the SDK with one command pip install mcp-agent-graph and directly load and run these graphs in Python. This means: developers can integrate into existing systems with code; teams can collaboratively design through visual interfaces and finally deploy to production environments through code; your agent graphs can seamlessly embed into existing Python projects with free combination. The SDK provides complete graph loading, execution, and monitoring capabilities, making agent graphs powerful weapons in your code toolbox.
1️⃣ From need to Agent（从需求到智能体）
这是一个非常棒的功能！AI 生成智能体。在之前，您可能需要通过编写大量的代码来设计智能体，或者在前端编排您的工作流，增加一个个节点。mcp-agent-graph 提供了一种方案：您只需要将您的需求写出来，系统将为您生成一个良好的图！
从需求到图，可能只需要3分钟，或者更少！图生成完毕，您的画布就会出现这张图（agent），您可以查看图的每一个节点，整体的流程，以及readme文件！点开每一个节点，您可以看到他选择了哪些工具（mcp server），提示词，模型，节点之间的上下文传递...如果不满足需求，那么您可以使用 AI 优化图 的功能，将您的更新需求再次告诉AI，他将为您调整流程、增加或删减节点、修改提示词或者工具调用。
2️⃣ AI生成MCP工具（从需求到MCP）
这是面向future的功能！通过自然语言描述创建定制的MCP工具。
传统的MCP工具开发需要丰富的编程知识和对MCP协议的理解。mcp-agent-graph打破了这一障碍：只需描述您希望工具做什么，AI将为您生成一个完整的、生产就绪的MCP工具！
从描述到部署只需几分钟！只需提供需求，系统将：自动生成符合MCP标准的Python代码、创建包含所有依赖项的适当虚拟环境、处理端口管理和冲突检测、自动将工具注册到您的系统配置中、提供全面的文档和使用示例。生成的工具可以立即在您的智能体中使用，或与社区分享。这使MCP工具开发民主化，让每个人都能使用，无论技术背景如何。
3️⃣ 图嵌套图（分层世界）
这是一个架构创新！构建"Agent中的Agent"的分层智能系统。以往的工作流往往是扁平化的，当系统变得复杂时，会变得难以管理和维护。mcp-agent-graph 引入了图嵌套图的概念：任何完整的图都可以作为另一个图中的单个节点使用！这创造了无限的可能性。
层次化设计，无限扩展能力！您可以先构建一个"文档分析"图，包含文档解析、内容提取、格式转换等节点。然后将这整个图封装成一个节点，在更大的"知识管理"图中使用。这种分层设计让您能够：构建可复用的智能体模块、管理复杂的大型系统、实现真正的模块化开发。每一层都有清晰的职责边界，既保持了系统的整体性，又具备了极强的可维护性。
4️⃣ 图转 MCP Server
这是一个标准化功能！将智能体图导出为标准的MCP服务。在AI工具生态中，不同平台和框架之间的互操作性一直是个挑战。mcp-agent-graph 提供图转mcp功能：一键将任何图导出为标准的MCP server Python 脚本！
构建一次，处处运行！导出的MCP server完全符合MCP协议标准，可以被Claude Desktop、Cline、cursor等其他AI应用或任何支持MCP的系统直接调用。您的智能体瞬间变成了可被广泛集成的工具。导出的脚本包含完整的依赖管理、配置文件、安装说明，接收者可以立即部署使用。这为智能体的标准化和生态化奠定了基础。
5️⃣ 可视化图编辑器
画布即代码！您只需要在可视化画布上拖拽节点、连接线条，就能构建出复杂的智能体工作流。所见即所得，设计即开发！每个节点都有丰富的配置选项，您可以直接在界面上设置提示词、选择模型、配置工具调用、定义输入输出关系。节点之间的连线清晰展示了数据流向和执行顺序，让复杂的逻辑变得一目了然。实时预览功能让您随时查看当前设计的执行效果。
6️⃣ 节点即 Agent
每个节点都是独立的智能体。图中的每个节点都拥有完整的Agent能力！每个节点都可以调用工具、处理复杂任务。微服务化的智能体架构，每个节点都是一个专家！您可以为每个节点配置专门的角色提示词，让它成为特定领域的专家。一个节点可以是数据分析师，另一个节点可以是内容创作者，第三个节点可以是决策制定者。它们通过MCP server获得强大的工具能力，如访问文件系统、联网搜索、执行计算等。节点之间通过上下文传递进行协作，形成强大的智能体团队。
7️⃣ 智能体交易与传递
这是一个生态功能！完整的智能体打包、分享和部署解决方案。在当前的AI开发环境中，分享一个完整的智能体系统往往需要复杂的环境配置、依赖安装和文档说明，这极大限制了智能体的传播和复用。mcp-agent-graph 提供了智能体的完整生命周期管理：将智能体系统及其所有依赖项打包成自包含的、可移植的单元。
一键打包，一键部署，智能体生态化！系统自动生成全面的README文档，详细说明智能体的功能、要求、使用方法。接收者无需了解复杂的技术细节，就能快速理解和部署您的智能体。这个功能为智能体市场交易、团队协作、开源分享提供了完整的解决方案。您可以轻松地：与同事分享专业工具、向客户交付定制解决方案、在开源社区贡献您的创作。
8️⃣ Python SDK 深度集成
这是一个双轮开发模式！前端可视化设计，后端代码执行的完美结合。mcp-agent-graph 通过 Python SDK 提供了：前端拖拽设计，后端代码运行！设计与开发的完美融合，既见即得又代码可控！
您可以在前端可视化界面中快速设计和调试智能体图，然后通过 pip install mcp-agent-graph 一行命令安装SDK，直接在Python中加载和运行这些图。这意味着：开发者可以用代码集成到现有系统；团队可以通过可视化界面协作设计，最终通过代码部署到生产环境；您的智能体图可以无缝嵌入到现有的Python项目中，自由组合。SDK提供了完整的图加载、执行、监控能力，让智能体图成为您代码工具箱中的强大武器。
V1.3.1
V1.1.0更新
项目初始版本
V1.0.0
✨ 核心特性
1️⃣ 基于图的 Agent 开发框架
提供直观的可视化环境，让您轻松设计和构建复杂的智能体系统。
2️⃣ 节点即 Agent
图中的每个节点都是一个独立的智能体，可以利用 MCP server 的工具能力完成特定任务。
3️⃣ 图嵌套图（分层世界）
支持将整个图作为另一个图中的节点使用，实现层次化的智能体架构，创建"世界中的世界"。
4️⃣ 图转 MCP Server
将任何图导出为标准的 MCP server Python 脚本，使其可作为独立工具被其他系统调用。