Stop Paying the
MCP Tax.

jl replaces jupyter-mcp-server with a single bash call — no schema overhead in every context window, direct API, stateful kernel. Built and battle-tested at AIGENSA Labs.

View on GitHub

The Token Tax

jupyter-mcp-server loads 16 tool schemas into every context window. On long sessions, that's thousands of wasted tokens per request — paying repeatedly for a schema that never changes.

Round-Trip Overhead

MCP requires multiple round-trips per operation: schema negotiation, tool call, response parsing. One `jl exec` replaces the whole chain with a direct Jupyter REST + WebSocket call.

Setup Complexity

SSH tunnels, token management, kernel lifecycle, session tracking — jl handles all of it once via config. Your agent just runs code.

Everything You Need. Nothing You Don't.

Token-EfficientStateful KernelSSH Tunnel ManagementDirect REST APIMIT Open SourceRemote Server Support

Try jl Today

MIT licensed, open source, Python ≥ 3.11. Install with uv and start running notebooks from Claude Code in minutes.

View on GitHub