grokbook 0.1.4

Interactive notebook server for learning computer science

grokbook

Interactive notebook server for learning computer science. Works like Jupyter — code cells, markdown, persistent IPython kernels — with a built-in MCP server so AI tutors can create and manage notebooks for you.

Hey, can you create a new project in Grokbook about LLMs and transformer architecture, and a first notebook to teach me about the attention mechanism in PyTorch? Go step by step, introduce the different components of attention, and use both markdown and code to give me a comprehensive intro.

Install

Requires Python 3.14+ and uv. uv will auto-install Python 3.14 if you don't have it.

Run instantly (recommended)

uvx grokbook

That's it. Opens the notebook UI on localhost:8080 and the MCP server on port 8081. A welcome notebook is created on first run.

If you cd into a project directory with a .venv first, grokbook auto-picks that .venv as the kernel so your project's libraries are available. Otherwise, you pick a kernel from the kernel: dropdown in the header (or paste any Python path).

Install permanently

uv tool install grokbook
grokbook                   # now on your PATH

From source (dev)

git clone https://github.com/marco-jeffrey/grokbook.git
cd grokbook
uv sync
uv run grokbook

Usage

uvx grokbook               # or just `grokbook` after uv tool install

Flags:

uvx grokbook --port 3000                      # different port
uvx grokbook --python /path/to/python         # explicit kernel interpreter
uvx grokbook --allow-code-execution           # enable execute tools for MCP

Kernel auto-detection

Grokbook picks a Python interpreter on startup in this order:

1. --python CLI flag (explicit override)
2. $VIRTUAL_ENV (if an env is active in your shell)
3. ./.venv/bin/python or ./venv/bin/python in the current directory
4. First uv-managed Python found via uv python list
5. Fallback to the Python grokbook is running under

If the chosen interpreter is missing ipykernel, grokbook installs it automatically via uv pip install --python <env> ipykernel.

You can also switch kernels per-notebook from the UI: click the kernel: dropdown in the header → pick any discovered env, or paste a custom path at the bottom. Custom paths persist in ~/.grokbook/custom_envs.json.

Remote access (Tailscale / LAN)

By default, grokbook binds to 127.0.0.1 (localhost only). To access from other machines:

grokbook serve --host 0.0.0.0

Both the notebook server and MCP server bind to all interfaces. Access from another machine at http://<ip>:8080.

Warning: Grokbook executes arbitrary Python code. Do not expose it to untrusted networks.

CLI reference

grokbook [OPTIONS]                # Start notebook + MCP servers (= grokbook serve)
grokbook serve [OPTIONS]          # (explicit form)
  --host TEXT                     # Bind address (default: 127.0.0.1)
  --port, -p INT                  # Notebook server port (default: 8080)
  --mcp-port INT                  # MCP server port (default: 8081)
  --python PATH                   # Python interpreter for kernels
  --db PATH                       # Database file (default: ~/.grokbook/grokbook.db)
  --allow-code-execution          # Enable execute/kernel tools in MCP

grokbook mcp [OPTIONS]            # MCP server standalone (stdio, for Claude Desktop)
  --allow-code-execution          # Enable execute/kernel tools

MCP Integration

On startup, grokbook prints an MCP config block you can paste directly into Claude Desktop or LM Studio:

{
  "mcpServers": {
    "grokbook": {
      "command": "uvx",
      "args": ["grokbook", "mcp", "--allow-code-execution"],
      "env": {
        "GROKBOOK_API_URL": "http://localhost:8080/api"
      }
    }
  }
}

Omit --allow-code-execution to restrict the MCP server to read/write operations only (no code execution).

The grokbook mcp command runs in stdio mode for Claude Desktop. For HTTP-based MCP clients (LM Studio, remote agents), the built-in MCP server on port 8081 is already running when you start grokbook serve.

Always available: list_notebooks, get_notebook, create_notebook, rename_notebook, duplicate_notebook, list_projects, create_project, rename_project, move_notebook, create_cell, insert_cell, read_cell, write_cell, delete_cell, move_cell, duplicate_cell, change_cell_type, clear_output, clear_all_outputs

With --allow-code-execution: execute_cell, run_all_cells, kernel_status, restart_kernel, get_variables, interrupt_kernel

To enable code execution via MCP:

grokbook serve --allow-code-execution

Features

• Code cells with streaming execution, rich output (images, HTML, SVG, pandas tables)
• Markdown cells with GitHub-flavored rendering
• Persistent IPython kernels — one per notebook, variables carry over between cells
• Per-notebook kernel picker — auto-discovers uv, .venv, and jupyter kernelspecs; install ipykernel into any env with one click
• Keyboard-driven — Vim-like command/edit modes (j/k, a/b, dd, Shift+Enter)
• Import/export Jupyter .ipynb files
• Variables inspector panel
• Dark/light theme, wide mode, autocomplete, signature tooltips
• Live sync across browser tabs via SSE

Keyboard Shortcuts

Grokbook uses two modes, inspired by Vim:

Command mode (press Escape to enter):

Key	Action
j / k	Navigate between cells
Enter	Edit selected cell
a / b	Insert cell above / below
m	Convert to markdown
y	Convert to code
dd	Delete cell
Cmd+Shift+Up/Down	Move cell up / down

Edit mode (press Enter or click a cell):

Key	Action
Shift+Enter	Execute cell, move to next
Cmd+Enter / Ctrl+Enter	Execute cell, stay in place
Escape	Back to command mode
Tab / Shift+Tab	Indent / dedent

Vim mode

Enable Vim keybindings from the editor settings panel (gear icon). When active:

• Full Vim motions in code cells (normal, insert, visual modes)
• jk is mapped to Escape in insert mode for quick mode switching
• Block cursor in normal mode, line cursor in insert mode

Architecture

Browser ──SSE──▶ Stario server (:8080) ──ZMQ──▶ IPython kernel
   │                  │
   │  Datastar        │  SQLite (~/.grokbook/grokbook.db)
   │  (reactive       │
   │   signals)       ├── REST API (/api)
   │                  │
   ▼                  ▼
 DOM patches      MCP server (:8081)
 via SSE          (FastMCP, for LLM agents)

License

MIT