Agents – getting started

Nodality is written by people and by coding agents. A page is two plain arrays – elements (what exists) and nodes (what is done to it) – and plain data is what a language model writes well and what a validator can check before anything renders.

This section is for the agent that writes pages. For an agent that operates a finished page, see Agent surface.

One command

npx nodality skill

It installs the /nodality skill – the authoring workflow, the node families, and the house rules the schema cannot encode – into Claude Code (.claude/skills/) and Cursor (.cursor/rules/), and registers the MCP server in the project's config.

Flag	Effect
--claude	Claude Code only
--cursor	Cursor only (per project)
--global	Install into your home directory instead of the project
--dir=<path>	Copy SKILL.md for any other agent
--no-mcp	Skip registering the MCP server

Re-run it after upgrading nodality to refresh the skill in place.

The MCP server on its own

{ "mcpServers": { "nodality": { "command": "npx", "args": ["nodality", "mcp"] } } }

What an agent has without a network

schema.json (every element type and parameter, generated from the source) and API.md (an index of these docs) ship inside the npm package, so an agent reading node_modules/nodality/ has the vocabulary to hand. llms.txt and llms-full.txt index the site.

The loop

1. Read the vocabulary: list_ops for nodes, get_schema for the element type you are about to write.
2. Write the pair.
3. validate_nodes – fix every error it reports.
4. preview – render it.
5. check_page – fix what the rendered page actually does wrong.

See Validation for what the reports look like.