Enterprise DNA
M MCP Servers Developer low

PsChina/deepseek-as-subagent

by Various

Run DeepSeek as a real sub-agent inside Claude Code / Codex CLI — DeepSeek gets its own 7-tool agent loop in a sandboxed workspace, not just a single LLM call.

P

MCP

PsChina/deepseek-as-subagent

Added 1 June 2026

#agent-loop #ai-agent #claude-code #codex-cli #deepseek #llm-orchestration #mcp #model-context-protocol

Overview

This project enables running DeepSeek as a real sub-agent inside Claude Code or Codex CLI. DeepSeek operates in its own agent loop with seven tools in a sandboxed workspace, not just a single LLM call.

Best for

Best for
Developers using Claude Code who want to delegate subtasks to a separate DeepSeek sub-agent.

Use cases

  • Integrating DeepSeek as a sub-agent within Claude Code sessions
  • Running multi-step tasks in a sandboxed environment alongside a primary agent
  • Leveraging DeepSeek's capabilities in a controlled sub-agent workspace

Notes

This project enables running DeepSeek as a real sub-agent inside Claude Code or Codex CLI. DeepSeek operates in its own agent loop with seven tools in a sandboxed workspace, not just a single LLM call.

0 stars on GitHub. Last updated 2026-06-01. Licensed MIT.

Use cases

  • Integrating DeepSeek as a sub-agent within Claude Code sessions
  • Running multi-step tasks in a sandboxed environment alongside a primary agent
  • Leveraging DeepSeek’s capabilities in a controlled sub-agent workspace

Pros

  • Gives DeepSeek its own full agent loop with multiple tools
  • Sandboxed workspace reduces cross-contamination between agents
  • Extends Claude Code/Codex CLI with another reasoning model

Cons

  • Dependent on Claude Code or Codex CLI as the host environment
  • No indication of performance tradeoffs or stability under heavy use
  • Limited to users already working within the Claude Code ecosystem

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.

Pros

  • Gives DeepSeek its own full agent loop with multiple tools
  • Sandboxed workspace reduces cross-contamination between agents
  • Extends Claude Code/Codex CLI with another reasoning model

Cons

  • Dependent on Claude Code or Codex CLI as the host environment
  • No indication of performance tradeoffs or stability under heavy use
  • Limited to users already working within the Claude Code ecosystem