Be a 10x Vibe Coder (Claude Code + Cursor + MCP)

Summary of Technological Concepts, Product Features, and Workflow Tips

• Goal of the episode: Teach an “AI coding workflow” for maximizing productivity with Claude Code and Cursor—including model selection, MCP usage, deployment/server notes, and multiple practical workflow “hacks.”

Core Workflow Approach: Use Claude Code + Cursor Simultaneously

• Run Claude Code in one pane/tab (with a terminal).
• Switch to Cursor to use its agent features.
• For iOS: edit code in Cursor, and Xcode reflects changes by opening the same folder/files.

Model Strategy (What to Pick and Why)

Claude Code

• Preferred model: Opus 4.1
• Constraint: usage is limited due to pricing/limits (historically more generous).
• Practical strategy:
  • Use Opus only for very complex problems.
  • Avoid exhausting the weekly limit.

Cursor

• Heavily relies on:
  • Plan Mode for reasoning/planning before execution
  • Planning model: “GPT 5.1 high”
  • Execution model: “Sonnet 4.7”
• Claimed counterintuitive benefit:
  • GPT-5.1 High with planning performs unusually well—possibly because it’s better at critical thinking/writing-style planning than a “coding-optimized” model.
• If Claude Code limits are reached:
  • Use Cursor Plan Mode for complex tasks.

How to Decide Between Claude Code vs Cursor

• Cursor Plan Mode
  • Preferred for very complex bug/fix scenarios
  • Produces better follow-up questions and more thorough planning
• Claude Code
  • Preferred for major architecture in few prompts
    • Example: “architect a whole app in 3–4 prompts”
  • Also described as stronger for UI-related work
    • Design, interactions, animations

Tutorial/Demo: Building a Polished UI Animation Quickly

• They scaffold an Xcode project, then use AI coding tools to create an animation similar to an AI “searching/calculating” sequence (Apple Notes-like behavior).

Workflow details shown:

• Use a very detailed dictated prompt to drive a staged UI flow:
  • “Searching” state with shimmering text
  • timed transitions:
    • “Analyzing”
    • “Found X sources” (with favicons/icons)
    • “Calculating”
  • finally show hard-coded calories to validate the animation behavior
• Use Cursor Plan Mode and Claude Code to compare outputs.

Animation debugging pro tip:

• In Xcode simulator, enable “slow animations” to spot issues more clearly.

Iteration/polish note:

• It took 10–20 prompts to reach the desired polish,
• but iteration via AI is still much faster than doing it manually.

Outcome insight:

• Even with similar Sonnet execution models, planning mode vs non-planning mode can still yield different quality results.

---

Rapid-Fire Workflow Tips

1. Use Plan Mode for every action in Cursor

   • Planning helps the model think longer and usually improves results (claimed ~20%+ output quality).

2. Claude Code keyword: “ultrathink”

   • Typing “ultrathink” makes Claude Code think harder (visual highlight/color change).
   • Reported minimal/no practical downside in token usage; used in most messages.

3. Claude Code background server tasks

   • Claude Code can run the server in the background and access server logs.
   • Helps debug crashes/timeouts without manually copying logs around.

4. MCP servers: what they’re used for (and which ones)

   • MCP hype is considered justified once you see practical coding workflow value.
   • Context 7 (Contacts 7)
     • Provides compressed, well-formatted latest documentation for LLM consumption.
     • Example: integrate PostHog using MCP docs (instead of URLs) so the model pulls and uses up-to-date docs reliably.
   • Supabase MCP
     • Automates/accelerates project setup (including configuration and security rules).
     • Security caution: theoretically could do destructive actions, so use carefully in production.
     • Recommendation: use it especially for correctness checks (security rules, indexes), even if you don’t fully trust it.
   • Mentions testing more MCPs later (e.g., Figma MCP).

5. PR/code review tools for solo developers

   • Tools like Bugbot (and another Claude Code-related one) can review changes via GitHub pull requests.
   • Purpose: catch security issues and bugs a solo developer can’t rely on humans to review.
   • Benefit: peace of mind, especially when security matters.

6. Dictation instead of typing

   • They “dictate everything” because it produces more detailed prompts and improves results.
   • Specifically use Whisper Flow, noting it understands developer terminology.

7. Use Claude Desktop “deep research” for technical decisions

   • When paying for Claude Code, Claude desktop/chat also gets additional usage.
   • Example workflow:
     • Run deep research like: “best way to store data in React”
     • Summarize documentation for implementation
   • Use the research to guide Claude Code/Cursor execution and improve correctness/pattern selection.

---

Tooling Recommendations for Mobile “Vibe Coding”

• For beginners/mobile apps:
  • Start with create anything (or similar tools like vzero/bolt) because:
    • better design quality
    • better instruction-following
• For more advanced workflows:
  • Move up to Claude Code/Cursor once limits are hit.

Other App Platform Opinions

• Tried vibe-code (Vibe Code) and ROR, but subjectively prefers Create Anything for:
  • design quality
  • instruction-following
• Belief: most of these tools likely use similar underlying models; differences are mostly:
  • UI/workflow
  • prompting behavior

---

Main Speakers / Sources

• Greg — podcast host
• Chris — guest; “vibe coder” and productivity app builder; documents workflow on YouTube
• Mentioned/credited sources (not primary speakers):
  • Sam Altman (OpenAI co-founder)
  • referenced podcasts/episodes by the guest
• Tools mentioned:
  • Claude Code, Cursor
  • MCP servers: Context 7 (Contacts 7), Supabase MCP
  • Bugbot
  • Whisper Flow