MCP, the Agent or the CLI
Chartnaut gives you three ways to have an AI write and run your scripts: an AI app you already use, connected over MCP; the Agent inside Chartnaut; and the CLI, driven by a coding agent such as Claude Code or Codex. All three work on the same account and the same scripts. They differ in where you work, what they can build and who pays for the AI.
Claude Code
CodexAI app over MCP | Agent | CLI | |
|---|---|---|---|
Where you work | The chat in Claude, ChatGPT, Cursor or Codex | Agent at the top of Chartnaut's sidebar, and each builder's chat | A terminal or a coding agent, with scripts as files in your folder |
What it builds | Indicators, definitions and studies | The same, plus Forward Insights, apps and views | Indicators, definitions and studies |
What it shows you | Numbers, events and study results in the chat | Each piece on your charts, with approval cards before it runs | Output in the terminal, and links that open a run in the app |
Who pays for the AI | Your plan with that app | AI credits, or your own Claude or ChatGPT subscription in Chartnaut Desktop | Your coding agent's plan. The CLI itself uses no AI |
Before it changes your account | The app asks you to confirm tool calls, as its own settings say | Approval cards: plan, data, study run and result | Nothing is saved until you or the agent run |
Setup | Add a connector and click Allow | None | Install the CLI, |
Counts against | AI credits and runs at once | API limits and runs at once | |
Plan | Starter and above | Starter and above | Starter and above |
Pick by the job
You are already in Claude or ChatGPT and want an answer. Ask there over MCP: "How often did BTC close above the prior day's high on 1-hour bars last quarter?" Nothing to install. Connect Chartnaut to an AI app
You want a study, a Forward Insight or an app, and to check each step on the chart. Use the Agent. It is the only one that works on your charts and builds pages. How the agent works
You want your scripts as files, under version control, edited in your own editor. Use the CLI with Claude Code or Codex. Working with Claude Code and Codex
Using more than one
They share everything, so you can move between them. A definition an AI app saves over MCP opens in the definition builder, chartnaut pull fetches it as files, and the Agent finds it when you ask for a study on top of it.
Saving is guarded the same way everywhere: a save names the version it started from, and if the script changed since, in the app or anywhere else, the save is refused with version_conflict and nothing is overwritten. Saving and versions
Claude Code can use Chartnaut both ways at once, through the MCP server and through the CLI. Claude compares the two there.