In last 6 months, I’ve coded 18 MVPs for clients using Cursor....
Prajwal Tomar@PrajwalTomar_
68 views
Apr 14, 2025
~3 min read
Advertisement
1
In last 6 months, I’ve coded 18 MVPs for clients using Cursor.
Here’s my full workflow:
→ Cursor Project Rules
→ Gemini Pro 2.5 for context
→ Sonnet 3.5 for execution
→ CodeGuide for docs
Bookmark this and copy my Cursor AI workflow: ↓
Here’s my full workflow:
→ Cursor Project Rules
→ Gemini Pro 2.5 for context
→ Sonnet 3.5 for execution
→ CodeGuide for docs
Bookmark this and copy my Cursor AI workflow: ↓
2
1/ Without planning, Cursor = Chaos
When it comes to coding with Cursor, context is everything. If you don't spend time planning, AI models will hallucinate
The result?
- Random folder structures
- Broken logic
- Hallucinated layouts
How I fix it:
I don’t rely on prompts. I Build a system that AI can follow.
When it comes to coding with Cursor, context is everything. If you don't spend time planning, AI models will hallucinate
The result?
- Random folder structures
- Broken logic
- Hallucinated layouts
How I fix it:
I don’t rely on prompts. I Build a system that AI can follow.
3
2/ Don’t use cursorrules file!
.cursorrules is a single file with global rules.
It works… until it doesn’t.
Problems:
- One-size-fits-all logic
- AI can’t the entire file every time
- Hallucinations when rules get too broad
AI needs structure. .mdc Project Rules fix that.
.cursorrules is a single file with global rules.
It works… until it doesn’t.
Problems:
- One-size-fits-all logic
- AI can’t the entire file every time
- Hallucinations when rules get too broad
AI needs structure. .mdc Project Rules fix that.
4
3/ The new standard: Project Rules
If Cursor doesn’t understand your product, it will hallucinate.
That’s why I generate 7 .mdc docs before I write a single line of code:
📁 .cursor/rules/
- backend_structure_document.mdc
- app_flow_document.mdc
- tech_stack_document.mdc
- frontend_guidelines_document.mdc
- backend_structure.mdc
- implementation_plan.mdc
- cursor_project_rules.mdc
These aren’t notes. They’re hard-coded rules for AI.
If Cursor doesn’t understand your product, it will hallucinate.
That’s why I generate 7 .mdc docs before I write a single line of code:
📁 .cursor/rules/
- backend_structure_document.mdc
- app_flow_document.mdc
- tech_stack_document.mdc
- frontend_guidelines_document.mdc
- backend_structure.mdc
- implementation_plan.mdc
- cursor_project_rules.mdc
These aren’t notes. They’re hard-coded rules for AI.
5
4/ Why this structure matters
Each .mdc file gives Cursor scoped knowledge:
- frontend_guidelines_document.mdc → styling, component rules
- backend_structure_document.mdc → API patterns, DB queries
- cursor_project_rules.mdc → global coding standards
- implementation_plan.mdc → step-by-step instructions
The result?
Cursor behaves like a junior dev, trained for your project.
Each .mdc file gives Cursor scoped knowledge:
- frontend_guidelines_document.mdc → styling, component rules
- backend_structure_document.mdc → API patterns, DB queries
- cursor_project_rules.mdc → global coding standards
- implementation_plan.mdc → step-by-step instructions
The result?
Cursor behaves like a junior dev, trained for your project.
6
5/ I generate these docs using @CodeGuidedev
It gives me:
- A complete PRD
- Detailed App Flow
- Tech Stack + API usage
- Design System (fonts, layout, spacing)
- Auth, DB, and backend setup
- A 50-step Implementation Plan
I save each as an .mdc file. Cursor reads them like a dev studying the codebase.
It gives me:
- A complete PRD
- Detailed App Flow
- Tech Stack + API usage
- Design System (fonts, layout, spacing)
- Auth, DB, and backend setup
- A 50-step Implementation Plan
I save each as an .mdc file. Cursor reads them like a dev studying the codebase.
0:47
7
6/ Use the right model for the right job
Here’s the exact setup I use after delivering 18 MVPs using Cursor:
Gemini 2.5 Pro
→ Scan full codebase (1M context)
→ Catch issues
→ Update .mdc docs
Claude Sonnet 3.5 / 3.7
→ Execute features
→ Fix logic
→ Build from the implementation plan
Gemini thinks. Sonnet builds.
Here’s the exact setup I use after delivering 18 MVPs using Cursor:
Gemini 2.5 Pro
→ Scan full codebase (1M context)
→ Catch issues
→ Update .mdc docs
Claude Sonnet 3.5 / 3.7
→ Execute features
→ Fix logic
→ Build from the implementation plan
Gemini thinks. Sonnet builds.
0:09
8
7/ Cursor Agent + Implementation Plan = fast builds
The implementation plan from CodeGuide is the blueprint.
I attach it as implementation_plan.mdc and prompt:
“Follow Step 1 from the plan.”
Cursor Agent builds step-by-step.
No assumptions. No jumping around. Just clean execution.
The implementation plan from CodeGuide is the blueprint.
I attach it as implementation_plan.mdc and prompt:
“Follow Step 1 from the plan.”
Cursor Agent builds step-by-step.
No assumptions. No jumping around. Just clean execution.
9
8/ Supabase + MCP handles the backend
Cursor can now:
- Connect to Supabase
- Create + modify tables
- Apply policies
- Sync local + remote DBs
This entire flow is triggered from your .mdc backend structure + schema docs.
Again, planned with CodeGuide. Executed by AI.
Cursor can now:
- Connect to Supabase
- Create + modify tables
- Apply policies
- Sync local + remote DBs
This entire flow is triggered from your .mdc backend structure + schema docs.
Again, planned with CodeGuide. Executed by AI.
10
9/ Final Setup = My 2025 Dev Stack
- CodeGuide → generates your AI Knowledge Base
- .cursor/rules/*.mdc → context boundary
- Gemini 2.5 → scan + update
- Sonnet 3.5/3.7 → execute + debug
- Cursor Agent → follow the plan
- Supabase MCP → automate backend
- Vercel → deploy in 1 click
This is how I ship MVPs in 5–6 weeks.
- CodeGuide → generates your AI Knowledge Base
- .cursor/rules/*.mdc → context boundary
- Gemini 2.5 → scan + update
- Sonnet 3.5/3.7 → execute + debug
- Cursor Agent → follow the plan
- Supabase MCP → automate backend
- Vercel → deploy in 1 click
This is how I ship MVPs in 5–6 weeks.
11
10/ Final takeaway
AI doesn’t replace developers.
It replaces chaos, if you give it structure.
- Planning > prompting.
- Context > guessing.
- Execution > exploration.
Use Project Rules. Use multiple models.
Use @CodeGuidedev to generate your docs.
And @cursor_ai will finally code like a real dev.
Bookmark this. It’ll save you weeks.
AI doesn’t replace developers.
It replaces chaos, if you give it structure.
- Planning > prompting.
- Context > guessing.
- Execution > exploration.
Use Project Rules. Use multiple models.
Use @CodeGuidedev to generate your docs.
And @cursor_ai will finally code like a real dev.
Bookmark this. It’ll save you weeks.