OpenCode Plugin: NotebookLM

Access Google NotebookLM from OpenCode AI coding assistant.

Features

• 8 tools with context inference
• Auto-auth via CDP - Chrome auto-launches when needed
• Notebook state persistence (auto-select active notebook)
• Multi-turn conversations
• Create & manage notebooks
• Add sources (URLs, text, Google Drive)
• Query AI about your sources
• Generate audio, reports, flashcards, infographics
• Deep research with web search

Installation

Option 1: npm package

bun add opencode-plugin-notebooklm

Add to opencode.json:

{
  "plugins": ["opencode-plugin-notebooklm"]
}

Option 2: Local plugin

Copy to .opencode/plugins/notebooklm/:

cp -r opencode-plugin-notebooklm .opencode/plugins/notebooklm

Authentication

Automatic (Recommended)

Plugin auto-launches Chrome when auth is needed:

1. Chrome opens with NotebookLM
2. Login to your Google account
3. Plugin extracts cookies automatically
4. Done! Chrome can be closed after login

First time setup: Just use any NotebookLM tool - Chrome will open automatically.

Manual (Fallback)

If auto-auth fails, save cookies manually:

save_auth_tokens({ cookies: "your-cookie-header-from-devtools" })

To get cookies:

1. Open https://notebooklm.google.com in Chrome
2. Open DevTools (F12) > Network tab
3. Refresh page, click any request
4. Copy the Cookie header value from Request Headers

Auth Recovery Flow

When auth expires, plugin attempts 4-layer recovery:

Auth Error → Refresh CSRF → Reload Disk → CDP Auto-refresh → Manual Auth

1. Refresh CSRF - Re-extract tokens from page
2. Reload Disk - Load cached tokens from ~/.notebooklm-mcp/auth.json
3. CDP Auto-refresh - Launch Chrome, extract fresh cookies
4. Manual Auth - Prompt for save_auth_tokens (last resort)

Tools Reference

Notebook Management (4 tools)

| Tool | Description | |------|-------------| | notebook_list | List all notebooks | | notebook_create | Create a new notebook | | notebook_get | Get notebook details + AI summary | | notebook_query | Ask AI about sources (multi-turn) |

Source Management (1 tool)

| Tool | Description | |------|-------------| | source_add | Add sources to notebook |

source_add parameters:

• urls - URL(s) separated by space/newline
• drive_id - Google Drive document ID
• text - Plain text content
• title - Title (required for text)
• notebook_id - Target notebook

Research & Studio (2 tools)

| Tool | Description | |------|-------------| | research_start | Start web research (fast/deep mode) | | studio_create | Generate content (audio/report/flashcards/etc) |

Auth (1 tool)

| Tool | Description | |------|-------------| | save_auth_tokens | Save cookies from browser (fallback) |

Skills

Optional workflow guides available in skills/:

| Skill | Description | |-------|-------------| | nlm-index | Index docs/repos to NotebookLM |

Use with: skill({ name: 'nlm-index' })

Smart Features

Context Inference

Notebook ID is auto-inferred when:

• Only one notebook exists (auto-selected)
• A notebook was recently accessed (session state)

# No need to specify notebook_id if context is set
notebook_query({ query: "What are the main topics?" })

Adding Sources

# Add URL(s)
source_add({ urls: "https://example.com/article" })
source_add({ urls: "https://example1.com https://example2.com" })

# Add Google Drive document
source_add({ drive_id: "1BxiMVs0XRA5nFMdKvBdBZjgmUUqptlbs" })

# Add text content (title required)
source_add({ text: "Some text...", title: "My Notes" })

Multi-turn Conversations

# First query
notebook_query({ query: "What is React?" })

# Follow-up (uses same conversation)
notebook_query({ query: "How about hooks?", conversation_id: "..." })

Studio Content Types

studio_create supports:

• audio - Audio overview/deep dive
• report - Briefing document
• flashcards - Study flashcards
• infographic - Visual infographic
• slide_deck - Presentation slides
• data_table - Structured data table

Usage Examples

Create notebook and add sources

> Create a notebook about React hooks
> Add the React docs: https://react.dev/reference/react/hooks

Query your sources

> Ask: "What are the best practices for useEffect cleanup?"

Generate content

> Generate an audio overview of my notebook
> Create flashcards for React hooks

Research

> Research "React Server Components best practices"

Development

# Install dependencies
bun install

# Build
bun run build

# Type check
bun run typecheck

# Test
bun test

Architecture

src/
├── index.ts              # 8 tools + hooks
├── errors.ts             # AppError with structured errors
├── config.ts             # Configuration
├── types.ts              # TypeScript types
├── auth/
│   ├── manager.ts        # AuthManager singleton (in-memory cache, recovery)
│   ├── cdp-provider.ts   # Chrome DevTools Protocol auth (auto-launch)
│   ├── cdp-resolver.ts   # Browser detection (Chrome, Edge, Brave)
│   └── cdp-browsers.ts   # Browser registry per OS
├── hooks/
│   └── index.ts          # OpenCode hooks (session events)
├── state/
│   ├── session.ts        # Session state (active notebook, conversation)
│   └── cache.ts          # TTL cache with auto-sweep
└── client/
    ├── index.ts          # NotebookLMClient (singleton with refresh mutex)
    ├── transport.ts      # RPC transport with 4-layer recovery
    ├── codec.ts          # Request/response encoding
    ├── encoding.ts       # Data encoding utilities
    ├── recovery.ts       # Error recovery strategies
    ├── conversations.ts  # Conversation persistence
    └── services/
        ├── notebook.ts   # Notebook CRUD operations
        ├── source.ts     # Source management
        ├── query.ts      # AI query operations
        ├── research.ts   # Web research
        └── studio.ts     # Content generation

Key Design Patterns

• Singleton client with refresh mutex (prevents auth stampede)
• 4-layer auth recovery: CSRF refresh → disk reload → CDP auto-launch → manual
• Bun native APIs: Bun.spawn, Bun.sleep for performance
• Service layer per domain (notebook, source, query, research, studio)
• Transport layer with retry/backoff and auth refresh
• State management: Session (in-memory) + Cache (TTL-based)
• Proactive auth: Token expiry check before requests

Requirements

• Bun 1.0+
• Google Chrome (for CDP auto-auth)
• macOS / Linux / Windows

License

MIT

Credits

Based on notebooklm-mcp by jacob-bd.