Getting started

What you are starting

Three pieces (usually cloned from the same monorepo or sibling repos):

1. RC-API — Express/TS HTTP API (/health, /tools/list-all, /tools/execute, rules endpoints, …). Listens on a configured port (often 3001 in dev).
2. ReCloud_app — MCP server entry (server_workflow.js after build): one MCP tool named recloud; each capability is a subcommand matching a catalog tool name.
3. Tray app — Electron (ReCloudTrayApp_v2) and/or Tauri build: stores RC API key, picks Cursor config dir, writes mcp.json (RC server), can write .cursor/rules/recloud-behavior.mdc.

Prerequisites

• Node.js (LTS) for RC-API and MCP.
• Cursor (or another MCP client).
• RC_API_KEY valid for your RC-API deployment (tray stores it; MCP gets it via mcp.json env).

Gemini (or other model keys) are configured on the RC-API side, not by replacing RC_API_KEY with a Gemini key.

Order of operations

1. Configure and start RC-API (see that package’s .env.example / README).
2. Build or copy the MCP bundle your tray expects under its resources (mcp_server_dist pattern).
3. Start the tray, enter the API key, select the folder that contains mcp.json (e.g. ~/.cursor).
4. Restart Cursor (or reload MCP) so the RC server entry is picked up.
5. In the agent, call the MCP tool recloud with JSON shaped as:

{
  "subcommand": "summarizeText",
  "parameters": { }
}

Exact parameters depend on the tool; manifest/help resources are exposed by the MCP server (recloud://tool_manifest pattern).

Tray status icon

Tray implementations may show API connectivity (e.g. polling /health). That is independent of MCP stdio—do not infer MCP health from arbitrary TCP ports.

Next

• configuration.md — keys and files on disk
• faq.md — MCP tool shape and common failures