OthmaneBlial/term_mcp_deepseek

A prototype implementation of an MCP-like server for terminal environments, using the DeepSeek API. It demonstrates Model Context Protocol (MCP) concepts by enabling AI-driven shell command execution and tool invocation via a web interface.

Source: GitHub Repository

Features

• Chat Interface:
  • Web-based chat client (built with Flask and Tailwind CSS) for interacting with the server.
• AI Integration:
  • Integrates with the DeepSeek API to generate responses.
  • AI can include special instructions (lines beginning with CMD:) to trigger shell command execution.
• Terminal Command Execution:
  • Executes shell commands in a persistent Bash session via pexpect.
  • Returns command output to the client.
• MCP Endpoints:
  • /mcp/list_tools: Lists available tools (e.g., write_to_terminal, read_terminal_output, send_control_character).
  • /mcp/call_tool: Invokes specific tool commands on the server.
• API Endpoints:
  • /chat: Accepts POST requests with a message, processes it via DeepSeek, executes commands, and returns responses.
• Open Source:
  • Licensed under the MIT License.
• Proof-of-Concept:
  • Not fully compliant with the official MCP standard (e.g., lacks JSON-RPC support, session management, and some security features).
• Future Improvements (Planned):
  • JSON-RPC protocol support.
  • Real-time output streaming (SSE or WebSockets).
  • User sessions, authentication, and enhanced security.
  • Modularization of code for better maintainability.

Requirements

• Python 3.8+
• pip
• Valid DeepSeek API key

Category

• Code Execution Automation / MCP Servers

Tags

• mcp
• terminal
• shell
• reference-implementation

Pricing

• The project is open-source and free to use under the MIT License.