Make Shit Work — The autonomous bridge between NotebookLM and your coding agent.
A kr8tiv Project
Built by kr8tiv while shipping Jarvis Life OS — a project so complex it forced us to find better ways to work with AI agents.
We discovered the manual copy-paste loop between NotebookLM and coding agents was tedious but incredibly effective. We got tired of being the middleman. MSW automates that workflow.
This is a side quest born from necessity. We're sharing it because if it's helping us ship Jarvis faster, it might help you too.
| Platform | Link |
|---|---|
| Website | jarvislife.io |
| Twitter/X | @Jarvis_lifeos |
| Token | View on DexScreener |
You: *copies error from Claude Code*
You: *pastes into NotebookLM*
You: *gets perfect grounded answer*
You: *pastes back into Claude Code*
You: *agent succeeds*
You: *repeats 50 times a day*
You: "there has to be a better way"
Meanwhile, NotebookLM suggests follow-up questions that most people ignore. Those suggestions are often exactly what you need — and nobody clicks them.
MSW creates an Auto-Conversation Engine that:
- Auto-expands topics — Clicks ALL suggested questions in NotebookLM, scores relevance, keeps going 10+ levels deep
- Bidirectional comms — Injects agent errors INTO NotebookLM, gets grounded answers, feeds them back
- Persistent knowledge — Compiles everything into markdown and git commits it
- Zero manual intervention — The whole loop runs autonomously
┌──────────────────────────────────────────────────────────────────┐
│ MSW MCP SERVER │
│ │
│ ┌────────────────────────────────────────────────────────┐ │
│ │ Auto-Conversation Engine │ │
│ │ │ │
│ │ 1. Detect suggested topics in NotebookLM │ │
│ │ 2. Score relevance (string-similarity, 0-100) │ │
│ │ 3. Click high-scoring topics automatically │ │
│ │ 4. Recurse until convergence (10+ levels) │ │
│ │ 5. Inject agent errors → get grounded answers │ │
│ │ 6. Compile & git commit all findings │ │
│ └────────────────────────────────────────────────────────┘ │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ GSD Planner │ │ Ralph Runner │ │Browser Driver│ │
│ └──────────────┘ └──────────────┘ └──────────────┘ │
│ │
└──────────────────────────────────────────────────────────────────┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────────┐ ┌──────────────┐
│ .msw/ │ │ NotebookLM │ │ Chrome + │
│ state │ │ (browser) │ │ Playwright │
└──────────┘ └──────────────┘ └──────────────┘
| Foundation | Role | Link |
|---|---|---|
| GSD Protocol | Spec-driven planning & verification | GitHub |
| Ralph Wiggum Loop | Continuous iteration until convergence | Anthropic |
| NotebookLM MCP | Agent-to-NotebookLM bridge | GitHub |
Windows (PowerShell):
git clone https://github.com/kr8tiv-ai/kr8tiv-MSW.git
cd kr8tiv-MSW
.\setup.ps1Linux/macOS (Bash):
git clone https://github.com/kr8tiv-ai/kr8tiv-MSW.git
cd kr8tiv-MSW
chmod +x setup.sh && ./setup.shThe setup script will:
- ✅ Install dependencies
- ✅ Build TypeScript
- ✅ Configure MCP for Claude Code/Cursor/Windsurf
- ✅ Set up authentication flow
Then restart your IDE and the MSW tools will appear automatically.
See SETUP.md for detailed installation and authentication instructions.
| Tool | What It Does |
|---|---|
msw_init |
Initialize .msw/ with Google authentication and config |
msw_discover |
Run project discovery and gather relevant docs/notebooks before research |
msw_status |
Check job progress, config, or run health check (msw doctor) |
msw_notebook_add |
Link a NotebookLM notebook URL to the project (optional source auto-upload) |
msw_upload_sources |
Upload discovered sources into a NotebookLM notebook |
msw_research |
Auto-conversation engine — extract deep knowledge from NotebookLM (requires discovery first) |
msw_plan |
Generate PRD grounded in research findings |
msw_execute |
Run Ralph loop with NotebookLM feedback injection |
msw_verify |
Validate implementation against requirements |
Current flow:
- Run
msw_init - Run
msw_discover - Optionally run
msw_notebook_add/msw_upload_sources - Run
msw_research - Continue with
msw_plan -> msw_execute -> msw_verify
| Document | Description |
|---|---|
| SETUP.md | Complete installation and authentication guide |
| TROUBLESHOOTING.md | Common issues and solutions |
| CONTRIBUTING.md | Development guidelines and PR process |
| Example | Description | Time |
|---|---|---|
| VPS Debugging | Fix bot crashes using NotebookLM knowledge | 10 min |
| Feature Planning | Research → PRD → Execute with GSD + Ralph | 20 min |
All design documents, research reports, and reference material live in docs/:
| Document | Description |
|---|---|
| MSW_PRD.md | Full product requirements |
| MSW_PROTOCOL_ARCHITECTURE.md | Technical architecture |
| Architectural Blueprint | Research report on system design |
| Guide: Implementing MSW | Step-by-step implementation guide |
| Ralph Loop & MCP | Advanced agentic orchestration |
| Strategic Playbook | Go-to-market strategy |
| GSD Ralph Prompt | GSD + Ralph integration prompt |
MSW would not exist without GSD (Get Shit Done) by @official_taches. The spec-driven approach, phase planning, and verification loops changed how we build software with AI agents.
| Resource | Link |
|---|---|
| GitHub | glittercowboy/get-shit-done |
| Twitter/X | @official_taches |
No more fuckery. No more endless loops. No more manual copy-paste.
Just make shit work.
MIT License
@kr8tivai · jarvislife.io · @Jarvis_lifeos