Use the Claude Code CLI as the engine.
The same binary you already run in a terminal, driven from a window that also has your diagrams, your dependency audit and your dashboards in it.
Select Claude Code in the engine picker and Lucid Train runs the `claude` binary from your PATH as a local child process, talking to it with `claude -p` and stream-json in both directions. It is the same CLI you would run in a terminal, signed in the same way, so there is no second subscription, no API key for the engine and no proxy in the middle. If the binary is missing, the picker shows the install command; if it is installed but not signed in, it shows `claude auth login`.
How the connection actually works
The harness starts `claude -p` and keeps it alive, sending prompts and reading events as newline-delimited JSON in both directions. A long-lived process rather than one invocation per turn matters because it is what lets a session keep its context and be resumed rather than replayed from the top on every message.
Whose credential is used
On this route Lucid Train never sees the credential. It looks for the binary on PATH, spawns it as a child process, and speaks its protocol. Authentication is whatever that CLI already did when you signed into it, so the entitlement stays between you and the vendor and nothing is proxied, replayed or re-signed.
What Lucid Train adds on top
Claude Code is a terminal program. Running it underneath this harness puts the same agent next to a diagram canvas whose output it can be handed as a specification, a dependency audit that can shell out to real scanners, and a Grafana connection. Lucid Train also registers itself as an MCP server for the session, so the diagram tools arrive as mcp__lucid__ tools and are pre-approved rather than prompting on first use.
The two engine lists
One thing worth separating, because the two lists genuinely differ: the engine that runs a coding tab and the provider that powers Design mode are chosen independently. Coding tabs accept Claude Code, Codex, Cursor CLI and OpenCode. Design mode accepts Claude Code and Codex only. Picking Cursor for your coding tab does not move Design onto Cursor, because there is no Cursor provider behind that canvas.
What you keep and what you lose
You keep the model quality and the tool loop, since it is the same process. You lose the parts of Lucid Train that only the bundled engine implements: connecting a new Grafana instance and adding connectors both require the Lucid engine, though once a Grafana is connected every engine can query it.
What the engine picker needs
| Value | |
|---|---|
| Binary on PATH | claude |
| Install | npm i -g @anthropic-ai/claude-code |
| Sign in | claude auth login |
| Transport | claude -p, stream-json both ways |
| Drives Design mode | Yes |
| API key needed by Lucid Train | None |
Questions
Related
- Run Lucid Train on the Codex CLIJSON-RPC to a local app-server, and the only engine besides Claude Code that can also drive the design canvas.
- Run Lucid Train on the Cursor CLICursor's headless agent, with the diagram canvas and the approval gate wrapped around it.
- Run Lucid Train on OpenCodeThe open-source option, and the only engine that talks HTTP rather than stdio.
- Run Lucid Train on OpenRouterOne key, many models, and the only route here that lets you split Plan, Edit and Vision across different providers.
Last updated