pod-marimo

The marimo candy of the OpenCharly candy library, as a standalone repo (kind-prefixed naming). It ships the marimo reactive notebook server — which also runs as an MCP server on the same port — with the GPU OSM analytics stack, the Apache Airflow Python deps, and the marimo-team/skills set for AI agents.

What it provides

Installs marimo (a reactive notebook server that also runs as an MCP server via --mcp on port 2718) plus its data stack into a pixi env at ${HOME}/.pixi/envs/default: GPU-accelerated OSM analytics deps (Polars-GPU/cuDF, geopandas, quackosm), Apache Airflow Python deps (the airflow layer ships no pixi env of its own), and the marimo-team/learn curriculum. The marimo-team/skills set is unpacked read-only at /opt/marimo-skills (MARIMO_SKILLS_DIR) for AI agents driving marimo over MCP, and a baked marimo.toml turns auto_instantiate on so cells run the moment the notebook URL opens.

Property	Value
Service	marimo (marimo edit --host 0.0.0.0 --port 2718 --no-token --headless --mcp --mcp-allow-remote /workspace, priority 30)
Port	2718 (notebook UI + MCP at /mcp/server)
Requires	layer-cuda, layer-supervisord, plugin-mcp (the out-of-process mcp: check verb)
Volume	workspace at /workspace
Env	NVIDIA_PYTHON_PROJECT=~/.pixi, LD_LIBRARY_PATH=/usr/lib64, MARIMO_SKILLS_DIR=/opt/marimo-skills
mcp_provide	marimo at http://{{.ContainerName}}:2718/mcp/server (http transport)

--mcp-allow-remote is mandatory because the server binds 0.0.0.0 (marimo refuses non-localhost MCP requests without it); --no-token keeps the editor token-free for the dev posture. Every claim is observable: the pixi-env binaries, the skills directory, the config line, and — at deploy scope — the live notebook UI plus the MCP endpoint on /mcp/server.

How to use it

charly box build marimo
charly config marimo
charly start marimo
# notebook UI:  http://localhost:2718
# MCP endpoint: http://localhost:2718/mcp/server

The candy's own check: steps assert the marimo and airflow pixi binaries, the unpacked skills directory, the baked auto_instantiate config, the OSM/MCP curriculum imports, the reachable port and HTTP 200, and the mcp: ping / list-tools surface.

Layout

• charly.yml — the marimo: candy entity (description, require, env, env_accept, mcp_provide, volume, service, plan).
• pixi.toml / pixi.lock — the Python environment (marimo, the OSM/GPU stack, Airflow deps).
• build.sh — the pixi-builder post-install (the three PyG native extensions).
• .github/workflows/tag-on-merge.yml — CalVer tag + CHANGELOG/ on merge.
• CHANGELOG/ — per-CalVer history.
• README.md — this user overview.

Related

• Owning skill: /charly-versa:marimo-layer — the marimo layer, its pixi environment, the supervisord service spec, and the rendering patterns.
• /charly-versa:marimo-mcp — marimo's built-in MCP server and its read-only tool catalog.
• /charly-versa:notebook-osm — the self-authoring OSM notebook this env drives.
• opencharly/charly — the charly CLI and image builder
• opencharly/opencharly — the umbrella