Make DeepSeek BETTER Than Sonnet With Two MCPs (COMPLETELY FREE)



AI Summary

Video Summary

  • Objective: Testing DeepSeek V3.1 with Sequential Thinking Context 7 MCPS to create a website.

Key Points:

  • Context: Previous successes with other models (e.g., Gemini 2.5 Flash, Sonic 3.7).
  • DeepSeek Setup:
    • Utilized OpenRouter with 685 parameters.
    • Introduced a new MCP called Sequential Thinking for step-by-step guidance.
    • Used a service-based website prompt as benchmark.
  • Initial Setup:
    • Gathered materials and adjusted prompts for DeepSeek.
    • Emphasized keeping tags and modular sections in content.
  • Execution:
    • In “act” mode after planning.
    • Initial output with some errors (404 pages, language issues).
    • Required final adjustments to ensure production readiness.
  • Cost and Performance:
    • Total cost during the process remained under 10 cents.
    • DeepSeek provided a near-complete website despite errors.
  • Results:
    • While some issues persisted (404 errors, missing elements), overall output was an impressive attempt, rivaling capabilities of more expensive models.
    • Conclusion: DeepSeek can effectively build websites at a low cost, marking a significant improvement in performance over previous attempts.

Final Remarks

  • Potential for using DeepSeek in conjunction with other models (e.g., planning with Claude). This opens avenues for more efficient website creation.
  • Recommendation: Consider DeepSeek for affordable web development options.