English | ไธญๆ
MCP Agent Graph is a Multi-Agent System built on the principles of Context Engineering. It integrates Sub-agent, Long-term Memory, MCP, Agent-based Workflow, and other capabilities. By integrating Context Engineering best practices into a visual development experience, MCP Agent Graph enables developers to rapidly build, test, and deploy complex multi-agent applications.
| Try Online | https://agent-graph.com/ |
| Invitation Code | TEAM-QI10IT |
| Documentation | https://keta1930.github.io/mcp-agent-graph/ |
โ ๏ธ Important Note: The models on the demo site do not have API keys configured. You will need to add your own API keys in Model Management to use the platform.
- Framework
- Deployment Guide
- Core Features
- Future Roadmap
- Frontend Feature Showcase
- Citation
- WeChat Group
๐ Detailed Installation Documentation: docs/first-steps/install.md
| Component | Requirement |
|---|---|
| Operating System | Linux, macOS, or Windows (requires WSL2) |
| Docker | Version 20.10+ with Docker Compose |
| Python | Version 3.11+ |
| Memory | Minimum 4GB (8GB recommended) |
| Storage | At least 10GB available space |
git clone https://github.com/keta1930/mcp-agent-graph.git
cd mcp-agent-graphcd docker/mag_services
cp .env.example .env
# Edit .env file to configure necessary parameters (see installation documentation)
docker-compose up -dService Addresses:
- MongoDB Express (Database Management): http://localhost:8081
- MinIO Console (File Storage): http://localhost:9011
Using uv (Recommended):
cd ../.. # Return to project root
uv sync
cd mag
uv run python main.pyUsing pip:
cd ../.. # Return to project root
pip install -r requirements.txt
cd mag
python main.pyRun in Background:
nohup python main.py > app.log 2>&1 &Open browser and visit: http://localhost:9999
Login Page (Admin login with username and password configured in .env):
Registration Page (New users can register with invitation code):
Other Access Endpoints:
- API Documentation: http://localhost:9999/docs
- Health Check: http://localhost:9999/health
If you need to modify frontend code:
cd frontend
npm install
npm run dev # Development server: http://localhost:5173
npm run build # Build production versionNote: The repository includes pre-built frontend files. This step is only needed when developing or customizing the frontend.
| Feature | Description | Documentation |
|---|---|---|
| Agent | AI entities with capabilities to understand goals, use tools, iterate optimization, maintain context and long-term memory, solving open-ended tasks through autonomous action execution | Agent Docs |
| Graph (Workflow) | Orchestrate multiple agents into structured workflows, defining execution flow through nodes and edges, suitable for predictable multi-stage tasks | Graph Docs |
| Model | Support for multiple LLM models (OpenAI compatible), flexible API Key configuration | Model Docs |
| Memory | Short-term memory maintains conversation context, long-term memory stores user preferences and Agent knowledge base across sessions | Memory Docs |
| Prompt Center | Centralized management of reusable Prompt templates, supporting categorization, import/export, and cross-project references | Prompt Docs |
| Feature | Description | Documentation |
|---|---|---|
| Visual Graph Editor | Frontend drag-and-drop workflow design, supporting linear, parallel, conditional, and nested graph types, WYSIWYG | Graph Docs |
| Subgraph Nesting | Use entire Graphs as single nodes for nesting, enabling modular, reusable, and hierarchical workflow construction | Subgraph Docs |
| Handoffs (Smart Routing) | Nodes dynamically select next execution node, supporting intelligent decisions, conditional branching, and iterative optimization loops | Handoffs Docs |
| Task (Scheduling) | Scheduled or periodic automatic Graph execution, supporting cron expressions, concurrent instances, and execution history tracking | Task Docs |
| Feature | Description | Documentation |
|---|---|---|
| MCP Protocol Integration | Connect external tools and data sources (databases, APIs, file systems, cloud services, etc.) through standardized protocol, connect once and use everywhere | MCP Docs |
| Built-in Tool Set | Provides resource creation (Agent Creator, Graph Designer, MCP Builder, Prompt Generator, Task Manager), collaboration (Sub-agent, File Tool), and query (Memory Tool, System Operations) system tools | Tools Docs |
| Python SDK | Install via pip install mcp-agent-graph, build and manage Agent systems using Python code |
PyPI Package |
| Feature | Description | Documentation |
|---|---|---|
| Team Collaboration | Admins create invitation codes, manage team members, assign role permissions (Super Admin, Admin, Regular User) | Team Management |
| Conversation Management | Support conversation history viewing, file attachment management, and session context maintenance | Quick Start |
๐ Complete Roadmap: docs/roadmap/index.md
The platform continues to evolve, bringing more powerful Agent capabilities and better collaboration experiences to users.
The following features are coming soon or actively under development:
| Feature | Core Value | Documentation |
|---|---|---|
| Multimodal Support | VLM gives Agents visual understanding capabilities | Details |
| Team Resource Sharing | Share Agents, workflows, and Prompts within teams | Details |
| Agent Skills | Progressive context engineering to improve efficiency and capabilities | Details |
These features are under continuous exploration and planning:
| Feature | Core Value | Documentation |
|---|---|---|
| External Agent API | Open Agents to external calls, building a service ecosystem | Details |
| User Analytics | Effect evaluation and team insights | Details |
Entry interface for starting conversations with Agents, supporting quick selection of preset Agents or creating new conversations.
Create, configure, and manage agents, set system prompts, tools, and model parameters.
Visual drag-and-drop workflow designer, supporting multiple node types and complex process orchestration.
Configure and manage multiple LLM models, set API Keys and model parameters.
View and configure built-in system tools, including resource creation and collaboration tools.
Manage MCP server connections, configure external tool and data source integrations.
Centrally manage reusable Prompt templates, supporting categorization and version control.
Manage uploaded files and attachments, supporting file preview and organization.
View and manage Agent's long-term memory and knowledge base.
If you find MCP Agent Graph helpful for your research or work, please consider citing it:
@misc{mcp_agent_graph_2025,
title = {mcp-agent-graph},
author = {Yan Yixin},
howpublished = {\url{https://github.com/keta1930/mcp-agent-graph}},
note = {Accessed: 2025-04-24},
year = {2025}
}For questions, suggestions, or collaboration inquiries, feel free to reach out:
๐ง Email: yandeheng1@gmail.com