Features

Structure before code.

Nodeon doesn't generate code. It produces the brief your agent builds from: MCP-native graph access, scoped prompts, spec files, dependency order.

What Nodeon does

Six capabilities, all reading the same graph. Each one is a view of the structure you drew, so nothing is configured twice.

MCP server

Your coding agent connects to Nodeon via MCP and pulls task context on demand, node by node, in dependency order. No copy-paste, no file wrangling.

Freebuild web UI

Build and edit your mindmap on a canvas. Drag nodes, draw dependencies, refine specs. AI-assisted editing for splitting nodes, tightening vague descriptions, and reordering.

CLAUDE.md and cursor rules

Nodeon generates .md docs your agent already reads: CLAUDE.md, cursor rules and spec sheets. They stay in sync with your graph because they're derived from it.

Agent write-back

Your agent marks nodes done, adds discovered subtasks, and flags blockers. The graph is live project state, not a static plan.

Dependency-aware ordering

Nodeon figures out what to build first. Screens depend on data models. Features depend on auth. The export is sequenced so your agent doesn't have to guess.

Scoped context per task

Each node gets only the context it needs: what it depends on and what depends on it. Not the whole graph. Computed at edit time, cached for instant reads.

Where it runs

Edit the graph wherever you are. Read it from wherever your agent already lives.

  • Freebuild web UI: Build and edit the graph on a canvas. Drag nodes, draw dependencies, refine specs.
  • Mobile app: Sketch a structure away from your desk. One graph, synced everywhere.
  • MCP server: Your agent connects and pulls task context on demand, node by node, in dependency order.
  • File export: CLAUDE.md, cursor rules and spec sheets written into your repo, for anything that reads plain text.

Why planning matters

Agents don't fail because the model is bad. They fail because the input is underspecified and out of order. The brief is the bottleneck.

Ad-hoc prompting

  • Input is a paragraph in a chat.
  • The agent only sees what you remember to mention.
  • Prompts go in whatever order you send them.
  • Scope is never defined upfront.
  • The agent can't write back. Chat is one way.

Prompt templates

  • Input is a copy-paste scaffold.
  • Static. The template doesn't know your project.
  • Partial ordering. Templates have a fixed order.
  • Scope is template-level, not project-level.
  • The agent can't write back. Templates are static.

Nodeon

  • Input is a visual graph with typed nodes.
  • Every node includes its connections.
  • Tasks are exported in dependency order.
  • Scope is set per node and validated across the graph.
  • The agent writes back: marks tasks done, adds subtasks, flags blockers.