MCP

Lucid Train works with the Model Context Protocol in both directions: it connects to servers, and it runs as one.

Transport: stdio only

Connections are stdio, meaning the server runs as a local child process with piped stdin and stdout. Remote HTTP and SSE servers appear in the Connectors tab's registry search but cannot be connected. That is a real limit rather than a configuration step you are missing; a remote server needs a local stdio bridge in front of it.

Running Lucid Train as a server

This exposes the diagram tools to any MCP client, including Claude Code, Codex and Cursor. Tools arrive namespaced, as mcp__lucid__emit_diagram and similar.

The distinguishing part is that the diagram is not placed by the language model. Your agent supplies a semantic graph of components, tiers and edges with no coordinates, and the ELK layout engine computes positions and orthogonal routing, so the output cannot come back with overlapping boxes.

Connectors

The Connectors tab has a small first-party registry, GitHub, Vercel, Railway and Grafana, and can install arbitrary stdio MCP servers alongside them. There is no first-party Jira or Confluence integration; you can add Atlassian's own stdio server yourself.

Adding a connector requires the bundled Lucid engine. Once added, the server is available whichever engine is running.

Audit what you install

An MCP server is a program launched with your agent's permissions, and the common install instruction fetches the latest version at every start. The Audit tab scans your MCP configuration for unpinned launches, remote and plaintext endpoints and credentials passed through the environment. It is worth running after adding a server rather than before deciding to.