README.md•6.47 kB
# Agent Forge - 智能体锻造工具 (AI Agent Forge Tool)
[English](#english) | [中文](#chinese)
<a name="chinese"></a>
## 中文版
Agent Forge 是一个智能体创建和管理平台，能够创建和管理具有特定性格特征的智能体，并模拟它们对问题的回答。通过Agent Forge MCP，你可以快速构建起一个类似于[CO-STORM](https://github.com/stanford-oval/storm)的多智能体协作研究项目。
### 功能特点
- 智能体锻造：创建具有特定性格特征的智能体
- 思维模拟：模拟智能体回答问题
- 完整管理：支持智能体的查询、列表、删除等操作
- 多轮对话：支持深度的多轮对话交互
- 自然语言处理：基于 DeepSeek API 的高级语言理解能力
### 系统要求
- Go 1.24.1 或更高版本
- DeepSeek API 密钥
### 安装
```bash
git clone https://github.com/HundunOnline/mcp-agent-forge.git
cd mcp-agent-forge && make build
```
### MCP 配置
```json
{
"mcpServers": {
"mcp-agent-forge": {
"command": "/path/to/mcp-agent-forge",
"env": {
"DEEPSEEK_API_KEY": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxx",
}
}
}
}
```
### 配置说明
#### Configuration
#### Environment Variables
| 变量名 | 描述 | 默认值 | 是否必需 |
|--------|------|---------|----------|
| `DEEPSEEK_API_KEY` | DeepSeek API 密钥 | - | 是 |
| `LOG_LEVEL` | 日志级别 (debug, info, warn, error) | info | 否 |
| `LOG_PATH` | 日志文件路径 | ./logs | 否 |
| `CONFIG_PATH` | 配置文件路径 | ./config/config.yaml | 否 |
| `PORT` | 服务端口号 | 8080 | 否 |
| `DEBUG` | 调试模式开关 | false | 否 |
### 使用方法
- `expert_personality_generation`: 创建新的智能体
- `agent_answer`: 模拟智能体回答问题
- `get_agent`: 获取智能体信息
- `list_agents`: 列出所有智能体
- `delete_agent`: 删除智能体
### 示例
#### 基本用法
```go
// 创建智能体
{
"name": "expert_personality_generation",
"arguments": {
"agent_name": "马斯克思维模型",
"core_traits": "系统思维,第一性原理,工程思维,风险管理,创新思维"
}
}
// 智能体回答
{
"name": "agent_answer",
"arguments": {
"agent_id": "your_agent_id",
"context": "如何看待特斯拉的发展策略？",
"planned_rounds": 3,
"current_round": 1,
"need_more_rounds": false
}
}
```
### 实际应用案例
我们在 Claude AI 中创建了一个示例应用，展示了如何使用 Agent Forge 创建和管理专家智能体：
[Claude AI Demo](https://claude.ai/share/00213269-9ab9-4923-9349-70d1492cd71e)
这个示例展示了：
- 如何创建具有特定专业背景的智能体
- 如何进行多轮对话交互
- 如何利用智能体的专业知识解决问题
- 如何管理和调整智能体的行为
### 贡献指南
欢迎提交 Pull Request 或创建 Issue 来帮助改进这个项目。我们特别欢迎以下方面的贡献：
- 新的智能体模型和特征
- 性能优化
- 文档改进
- Bug 修复
- 新功能建议
### 许可证
本项目采用 MIT 许可证。详见 [LICENSE](LICENSE) 文件。
---
<a name="english"></a>
## English Version
Agent Forge is a platform for creating and managing AI agents with specific personality traits and simulating their responses to questions. Through agent forge mcp, you can quickly build a multi-agent collaboration research project similar to [CO-STORM](https://github.com/stanford-oval/storm).
### Features
- Agent Forging: Create agents with specific personality traits
- Thought Simulation: Simulate agent responses to questions
- Complete Management: Support for agent querying, listing, deletion, and other operations
- Multi-round Dialogue: Support for deep multi-round conversation interactions
- Natural Language Processing: Advanced language understanding capabilities based on DeepSeek API
### System Requirements
- Go 1.24.1 or higher
- DeepSeek API key
### Installation
```bash
git clone https://github.com/HundunOnline/mcp-agent-forge.git
cd agent-forge && make build
```
### MCP Configuration
```json
{
"mcpServers": {
"mcp-agent-forge": {
"command": "/path/to/mcp-agent-forge",
"env": {
"DEEPSEEK_API_KEY": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxx",
}
}
}
}
```
### Configuration Guide
#### Configuration
#### Environment Variables
| Variable Name | Description | Default Value | Required |
|--------------|-------------|---------------|----------|
| `DEEPSEEK_API_KEY` | DeepSeek API Key | - | Yes |
| `LOG_LEVEL` | Logging level (debug, info, warn, error) | info | No |
| `LOG_PATH` | Log file path | ./logs | No |
| `CONFIG_PATH` | Configuration file path | ./config/config.yaml | No |
| `PORT` | Service port | 8080 | No |
| `DEBUG` | Debug mode switch | false | No |
### Usage
- `expert_personality_generation`: Create a new agent
- `agent_answer`: Simulate agent responses
- `get_agent`: Get agent information
- `list_agents`: List all agents
- `delete_agent`: Delete an agent
### Examples
#### Basic Usage
```go
// Create an agent
{
"name": "expert_personality_generation",
"arguments": {
"agent_name": "Elon Musk Thinking Model",
"core_traits": "Systems Thinking,First Principles,Engineering Mindset,Risk Management,Innovation"
}
}
// Agent response
{
"name": "agent_answer",
"arguments": {
"agent_id": "your_agent_id",
"context": "What's your view on Tesla's development strategy?",
"planned_rounds": 3,
"current_round": 1,
"need_more_rounds": false
}
}
```
### Real Application Case
We created a sample application in Claude AI that demonstrates how to use Agent Forge to create and manage expert agents:
[Claude AI Demo](https://claude.ai/share/00213269-9ab9-4923-9349-70d1492cd71e)
This example shows:
- How to create agents with specific professional backgrounds
- How to conduct multi-round dialogue interactions
- How to utilize agents' expertise to solve problems
- How to manage and adjust agent behavior
### Contributing
We welcome Pull Requests or Issues to help improve this project. We especially welcome contributions in the following areas:
- New agent models and traits
- Performance optimizations
- Documentation improvements
- Bug fixes
- New feature suggestions
### License
This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.