RetireGolden MCP

Headless Model Context Protocol server for the @retiregolden/engine retirement-planning calculator.

Run it locally over stdio, connect your AI client (Claude Desktop, Cursor, etc.), and call typed tools: build/validate a plan, run projections, Monte Carlo, batch policy evaluation, and optimization — with session state held in memory (no disk required).

npm: @retiregolden/mcp
License: AGPL-3.0-only (server); MIT (skills under skills/)

RetireGolden Pro ships this same package pre-wired in the desktop app for out-of-the-box convenience. Anyone can install and run the public package.

Disclaimer

Educational / decision-support only — not tax, legal, or financial advice. See DISCLAIMER.md.

Related MCP server: projectionlab-mcp

Quick start

npx @retiregolden/mcp
# or
npm install -g @retiregolden/mcp
retiregolden-mcp

Configure your MCP client to launch retiregolden-mcp (or npx @retiregolden/mcp) over stdio.

Example Claude Desktop / Cursor snippet

{
  "mcpServers": {
    "retiregolden": {
      "command": "npx",
      "args": ["-y", "@retiregolden/mcp"]
    }
  }
}

Connect your AI client

Per-client, copy-paste setup for Claude Desktop, Claude Code, Cursor, and Codex (MCP config + agent skill / rules) is in docs/clients.md.

Tools (v1)

Read-only with respect to your finances and disk: no trades, no persistence. Session state (e.g. build_plan, clear_session) is held in memory only.

Agent skill

See skills/retiregolden/SKILL.md (MIT). Bench and product runners should pin the skill file digest used for scored runs.

npm publish

Releases are tag-driven: pushing a mcp-v<version> tag triggers the publish workflow (see .github/workflows/publish-mcp.yml).

To run locally after npm run build:

node bin/retiregolden-mcp.js

Related

• Engine: https://github.com/RetireGolden/RetireGolden/tree/main/packages/engine

• Benchmark harness: https://github.com/RetireGolden/RetireBench

• Free planner (no MCP — browser only): https://retiregolden.app/

Trademark

See TRADEMARKS.md. Forks must rename.