概要
What is Savvly MCP Server?
The server lets financial advisors, planning tools, and AI agents query Savvly’s longevity-linked investment fund data, model payout projections, and compare Savvly against alternative retirement products in real time — without manual lookup. It integrates with the Savvly 80+ fund, which invests pooled contributions in a low-cost S&P 500 ETF and makes milestone cash payouts at ages 80, 85, 90, and 95.
How to use Savvly MCP Server?
Install via one of five package types: npx @savvly/mcp-server, uvx savvly-mcp-server, dnx Savvly.McpServer, docker run -i --rm ghcr.io/savvly/savvly-mcp-server, or point a remote‑MCP host at https://api.savvly.com/mcp. For stdio clients (Claude Desktop, Cursor, Windsurf), use the command / args config shown in the README.
Key features of Savvly MCP Server
- Project payouts to ages 80, 85, 90, and 95
- Compare Savvly vs annuities and target‑date funds
- Check age, residency, and channel eligibility
- Search an audience‑tagged Q&A library
- Generate interactive chart widgets (supported hosts)
- Fully public; no account or credentials required
Use cases of Savvly MCP Server
- A financial advisor asks a conversational AI to compare Savvly with a client’s current annuity.
- A retirement planning tool calls the server to model a client’s full retirement trajectory with and without Savvly.
- An AI agent checks eligibility for a prospect and returns a payout projection.
- A user searches the Q&A library for audience‑specific FAQs about the Savvly fund.
FAQ from Savvly MCP Server
What does the server compare Savvly against?
The compare_savvly_vs_alternative tool can compare Savvly to annuities, target‑date funds, and other retirement products.
What runtime or dependencies are needed?
Any runtime that supports npm, PyPI, NuGet, or Docker, or simply a browser/client that can call the hosted remote endpoint at https://api.savvly.com/mcp.
Does the server collect any personal information?
No. The server is public and unauthenticated. It collects no personal information and no end‑user identifier. Hosted usage logs anonymous analytics (tool name and non‑identifying numeric inputs) via MCPcat; the local stdio variant sends no analytics.
What transports and authentication are supported?
Stdio for local clients (via npx, uvx, dnx, or docker) and streamable‑http for remote access. There is no authentication — the server is fully open.
Are projection results interactive?
On hosts that support MCP Apps, projection tools emit interactive chart widgets showing milestone payouts and retirement trajectories.
