Adding MCP servers to Lucid Train.
Registry search or a command and some environment variables. Anything that speaks stdio.
Lucid Train's Connectors tab adds MCP servers two ways: searching the official MCP Registry, or entering a command, arguments and environment variables manually. Any stdio server works, including Atlassian's official one. Four provider connectors ship built in: GitHub, Vercel, Railway and Grafana, where Grafana also brings Loki and Prometheus querying through its own MCP server.
Two ways to add one
Discover queries the official MCP Registry and lists what it finds, with a fallback to web search. Manual takes a name, a command, its arguments and any environment variables as key-value rows, which is the path for anything not in the registry or anything internal. Configuration lands in your config file with restricted permissions.
What ships built in
GitHub, Vercel and Railway are auth-status connectors that wrap the vendor CLI, detecting whether it is installed and who is logged in, and delegating login. Grafana is the deep one: it downloads the Grafana MCP binary, stores credentials in the OS keychain, and registers it, which is how Loki log queries and Prometheus metric queries reach the agent.
Servers reach the other CLIs too
Whatever you configure is re-rendered into the native format each execution engine expects, so the same connectors are available whether a turn runs on Lucid Train's own agent, Claude Code, Codex or Cursor. You configure once rather than per CLI.
Transport
Connections are stdio: 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, and 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.
Atlassian, stated precisely
There is no Jira or Confluence specific code in Lucid Train. What is true today is that the manual form accepts any command, so Atlassian's official MCP server can be added the same as any other stdio server, provided it runs locally. First-party Jira and Confluence integration, with the setup and auth handling that implies, is on the roadmap and is not shipping.
What is built in versus what you add
| Connector | How it works |
|---|---|
| GitHub | Built in, wraps the gh CLI for auth status and login |
| Vercel | Built in, wraps the vercel CLI |
| Railway | Built in, wraps the railway CLI |
| Grafana | Built in, downloads and registers the Grafana MCP server |
| Loki and Prometheus | Through the Grafana server's tools |
| Atlassian, Linear, Notion, others | Add the vendor's stdio MCP server yourself |
| Internal or custom servers | Manual: command, args, environment |
Questions
Related
- Architecture Diagram MCP ServerPoint your existing agent at it and get diagrams that a solver laid out rather than a language model.
- Architecture Diagrams in Claude CodeKeep working in Claude Code and get a diagram that a layout solver produced.
- Architecture Diagrams in CursorStay in the editor, get a diagram a solver arranged.
- draw.io MCP Server AlternativeThe difference is who decides where the boxes go.