PrismAgent

PrismAgent is a highly sophisticated, multi-model AI agent framework built on the Model Context Protocol (MCP). Like a prism that decomposes light into its constituent colors, PrismAgent decomposes complex user requests into specialized tasks, leveraging a suite of dedicated MCP servers to provide precise answers and execute actions.

🌟 Key Features

• Multi-Model Support: Seamlessly switch between state-of-the-art models like DeepSeek and Qwen.
• MCP Native: Fully utilizes the Model Context Protocol for tool discovery and invocation.
• Specialized Toolkits:
  • 🔢 Math Server: Advanced symbolic mathematics using SymPy (Calculus, Algebra, Statistics).
  • 🌐 Web Server: Comprehensive search capabilities (Text, Lens, Scholar, News) via Serper.dev and webpage content fetching.
  • 💻 Code Server: Secure, persistent terminal sessions and sandboxed file management.
• Extensible Architecture: Easily add new MCP servers to expand the agent's capabilities.
• Token Tracking: Built-in monitoring for API usage and costs.

🏗 Project Structure

PrismAgent/
├── main.py              # Main entry point for the agent
├── models/              # Model handlers (DeepSeek, Qwen, etc.)
├── mcp_servers/         # Specialized MCP tool servers
│   ├── math_server/     # Symbolic math and computation
│   ├── web_server/      # Web search and scraping
│   └── code_server/     # Terminal and file sandbox
└── config/              # Configuration and memory logs

🚀 Getting Started

Prerequisites

• Python 3.12+
• Serper.dev API Key (for web search)
• DeepSeek or Qwen API credentials

Installation

1. Clone the repository and navigate to the project directory:

   cd PrismAgent

2. Install dependencies:

   pip install -r requirements.txt
   # or using uv
   uv sync

3. Setup environment variables:

   # Windows
   $env:SERPER_SEARCH_API_KEY = "your_key_here"
   $env:DEEPSEEK_API_KEY = "your_key_here"

Running the Agent

You can interact with PrismAgent via the command line:

python main.py --model deepseek --model_name deepseek-chat --question "求解方程 x^2 + 5x + 6 = 0 的根，并搜索这个方程在物理学中的应用。"

Command Line Arguments:

• --model: Choose the provider (deepseek, qwen).
• --model_name: Specific model ID (e.g., deepseek-reasoner).
• --question: Your query or task.
• --mcp_servers: Choose which tools to load (default: math web code).

🛠 MCP Servers

Math Server

Provides high-level mathematical tools including symbolic differentiation, integration, equation solving, and statistical analysis.

Web Server

Empowers the agent with real-time information. Supports standard text search, academic search (Scholar), visual search (Lens), and full webpage content extraction.

Code Server

Allows the agent to write and execute code in a persistent environment. It supports file management and terminal command execution with built-in security auditing.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.