CLI Interface¶

Stargazer does not ship a custom terminal UI. CLI users connect to the MCP server using whichever MCP-compatible client they prefer.

The MCP server (stargazer serve) is the single interface between any frontend and the Python backend. Building a custom TUI would duplicate what battle-tested tools already provide — streaming, tool call rendering, input handling — with no domain-specific value.

Supported Clients¶

Any MCP host that supports stdio or streamable HTTP transport:

Client	Transport	Notes
Claude Code	stdio	`stargazer serve` as MCP server in project config
OpenCode	stdio	Configure in `.opencode/`
Claude Desktop	stdio	Add to `claude_desktop_config.json`
Cursor / Windsurf	stdio	IDE MCP server configuration
MCP Inspector	stdio	Development and debugging

Setup¶

stdio (most clients)¶

The client spawns stargazer serve as a subprocess and communicates over stdin/stdout.

{
  "mcpServers": {
    "stargazer": {
      "command": "stargazer",
      "args": ["serve"]
    }
  }
}

Streamable HTTP (remote access)¶

stargazer serve --http --port 8080

Flyte CLI¶

All Stargazer tasks and workflows are standard Flyte v2 entities and can be managed directly through the Flyte CLI without the MCP server. This includes registering, running, and inspecting tasks and workflows.

For a local TUI to browse and launch tasks interactively:

flyte start tui

See the Flyte CLI reference for the full command set.

What Users Get¶

Regardless of client, users have access to all bioinformatics task tools, composite workflow tools, resources for context, prompt templates, and mode-aware tool registration. The MCP server handles mode detection, type serialization, and tool registration. The client handles LLM interaction and rendering.