All Webinars

Cursor

A live workshop covering the most underused Cursor features - from debug mode and the built-in browser to workspaces, skills, MCP servers, and model selection strategies.

Cursor Power User Workshop
Key Takeaways
  • 1 Use Sonnet 4.5 or Composer 1.5 for simple UI tasks, and Opus 4.6 or Codex 5.3 for complex backend debugging - picking the right model per task saves money
  • 2 Open multiple repos in one Cursor workspace (File > Add Folder to Workspace) so the agent has context across frontend and backend simultaneously
  • 3 Use plan mode for any complex or backend task - it improves results by 20-30% by letting the model think before executing
  • 4 Run your dev servers inside Cursor terminals so the agent can see logs directly - no more copy-pasting error output
  • 5 Rename terminal tabs (e.g. 'backend', 'frontend') so the Cursor agent picks the right log context automatically
  • 6 Hook up read-only MCP servers to your database (Postgres, Firebase, Supabase) so the agent can query schemas and user data while debugging
  • 7 Use Cursor Browser to let the agent visually see and interact with your app - it can click elements, fill forms, and read console logs
  • 8 Use the element selector in Cursor Browser to tag specific UI elements instead of screenshotting - it gives the agent more precise context
  • 9 Use debug mode for backend issues - it auto-inserts instrumentation, extracts only its own logs, gives reproduction steps, then cleans up after itself
  • 10 Dictate your prompts instead of typing - you'll write more detailed instructions, which dramatically improves one-shot accuracy and saves tokens
  • 11 Migrate heavy cursor rules into on-demand skills so they only load when needed - keeps the context window clean and improves performance
  • 12 Start a new chat after ~10 messages if the agent isn't converging - the context window is precious and quality drops as it fills up
Transcript
Model Selection Strategy

The four recommended models: Sonnet 4.5 and Composer 1.5 for cheaper UI tasks, Opus 4.6 and Codex 5.3 for complex backend work. Using a more expensive model that solves it in one prompt is often cheaper than 5-10 attempts with a cheap model.

Workspaces - Multi-Repo Context

Opening multiple code repositories in a single Cursor workspace via File > Add Folder to Workspace. The agent can see both frontend and backend code simultaneously, critical for debugging cross-repo issues.

Plan Mode & Agent Layout

Toggling between agent mode, plan mode, debug mode, and ask mode. Plan mode makes the agent think through scenarios before executing, yielding 20-30% better results on complex tasks.

Terminal Management & Log Visibility

Opening and renaming terminal tabs in Cursor. The agent has context of terminal windows and can read logs directly. Tag specific named terminals in prompts to point the agent to the right logs.

MCP Servers for Database Debugging

Connecting read-only MCP servers to databases (Postgres, Firebase, Supabase, AWS). The agent can run SQL queries, inspect schemas, and investigate specific user accounts.

BugBot - AI PR Reviews

Cursor's BugBot ($40/month) for automated PR reviews across 20+ repos. Gives targeted recommendations with a fix-in-Cursor button that opens the IDE with full context.

Dictation for Better Prompts

Using WhisperFlow for voice dictation. Dictating lets you give highly detailed descriptions which dramatically improves one-shot accuracy and reduces token spend.

Cursor Browser Deep Dive

Cursor's built-in browser lets the agent physically see your app. Covers manual resizing for mobile testing, the element selector for targeting UI components, and browser console access.

Cursor Skills vs. Rules

Skills are like rules but load on-demand, keeping the context window clean. Migrate non-essential rules to skills using the built-in migrate command.

Running Claude Code Inside Cursor

Running Claude Code in Cursor's terminal panel with a split view. Claude Code in the terminal has context of open files. You can spin up multiple instances as separate tabs.

Claude Code vs. Cursor - Cost Analysis

Cursor's $200 Ultra plan gives ~$400 in API credits. Claude Code's $200 plan gives ~$2,400 in credits (12x). From a pure cost perspective, Claude Code is the better deal.

Debug Mode - The Hidden Weapon

The most underused Cursor feature. It automatically inserts targeted debug logs, extracts only its own output, provides reproduction steps, monitors while you reproduce, then cleans up all instrumentation.

Q&A - Front-End Design, Figma, Tools

Screenshot fidelity tips, why Figma MCP isn't worth it, context window management, plan persistence with memory commands, hardware recommendations, and alternative tools like Ghostty and Dia browser.