About QuantCalc Retirement Engine
QuantCalc Retirement Engine is a hosted calculation service for Claude users assessing retirement plans and tax-aware withdrawals. Its read-only tools run retirement projections, compare return assumptions, explain methodology, and estimate a 2026 tax return.
Key features
- 01Run retirement success projections
- 02Model tax-aware withdrawals and Roth conversions
- 03Compare published return assumptions
- 04Inspect modeled methods and exclusions
- 05Estimate a 2026 federal and state tax return
Use cases
- 01Evaluate retirement-plan success rates
- 02Compare withdrawal orders and Roth-conversion plans
- 03Review results under different return assumptions
- 04Estimate a 2026 tax refund or amount owed
MCP support
The hosted engine connects to Claude through a plugin or custom connector over Streamable HTTP, and exposes six read-only calculation tools. The 2026 tax-return estimate requires a QuantCalc PRO or Tax Season Pass licence key.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://mcp.quantcalc.appclaude mcp add --transport http app-quantcalc-retirement-engine https://mcp.quantcalc.app{"mcpServers":{"app-quantcalc-retirement-engine":{"type":"http","url":"https://mcp.quantcalc.app"}}}Open in Cursor{"servers":{"app-quantcalc-retirement-engine":{"type":"http","url":"https://mcp.quantcalc.app"}}}Open in VS Code{"mcpServers":{"app-quantcalc-retirement-engine":{"type":"http","url":"https://mcp.quantcalc.app"}}}{"mcpServers":{"app-quantcalc-retirement-engine":{"serverUrl":"https://mcp.quantcalc.app"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
7| Parameter | Type | Required | Description |
|---|---|---|---|
| pension | number | No | Yearly pension income. Default 0. |
| allocations | array | No | Percentages summing to 100: [US stocks, international stocks, bonds, real estate, cash]. Default [60,10,25,5,0]. |
| current_age | integer | Yes | Current age of the primary person. |
| bucket_years | integer | No | Only with withdrawal_strategy bucket: years of spending held in the cash bucket, 1-10. |
| return_model | enum | No | How returns are generated. parametric (default): the forward-looking published assumptions. bootstrap: stationary block bootstrap of historical months. historical: replays real months from 1926 in order. The last two need a PRO licence; without one the engine runs parametric and the result says so.parametric | bootstrap | historical |
| ss_start_age | integer | No | Age Social Security starts. Default 67. |
| inflation_rate | number | No | Annual inflation as a percent, e.g. 2.5. Default 2.5. |
| retirement_age | integer | No | Age work income stops. Defaults to current age. |
| returns_source | string | No | Which published return set to use (see list_return_assumption_sources). Default jpmorgan. |
| annual_spending | number | Yes | Planned yearly spending in today's dollars, before income tax (this tool runs the standard engine, which does not model tax). |
| current_savings | number | Yes | Total invested portfolio today, in dollars. |
| life_expectancy | integer | No | Age the plan must last until. Default 92. |
| social_security | number | No | Yearly Social Security in today's dollars. Default 0 — and 0 is reported as an explicit assumption, not hidden. |
| pension_start_age | integer | No | Age the pension starts. Default 65. |
| withdrawal_strategy | enum | No | Spending rule. fixed (default): the planned amount, inflation-adjusted, every year. guyton-klinger: guardrails that cut or raise spending as the portfolio moves. vpw: variable percentage withdrawal. bucket: a cash reserve of bucket_years of spending.fixed | guyton-klinger | vpw | bucket |
| monthly_contribution | number | No | Monthly savings until retirement. Default 0. |
Details
From the project READMEQuantCalc Retirement Engine — Claude plugin
Gives Claude the retirement engine behind quantcalc.app, so a projection is computed rather than estimated.
A language model cannot run thousands of correlated return paths against a spending plan and get the odds right. It can ask something that can.
Privacy policy · Terms · Support: [email protected] · Documentation
Install
Install from the Claude plugin directory. The plugin installs switched off, and you enable it when you want it. Step-by-step instructions are in SETUP.md. To use the server without the plugin, add it as a custom connector:
https://mcp.quantcalc.app
What's in the plugin
- Connector: the hosted QuantCalc engine over streamable HTTP. No account or sign-in is needed.
- Skill
retirement-projections: loads when a plan, withdrawal rate or success rate comes up. It keeps the assumptions attached to every number and flags differences that fall inside the margin of error. - Skill
setup: connection steps, what a licence key changes, and troubleshooting.
Tools
| Tool | Returns |
|---|---|
run_retirement_projection | Success rate, ending-portfolio distribution, and the assumptions that produced them |
run_tax_aware_projection | The recommended withdrawal order and Roth-conversion programme, its present-value advantage and range against traditional-first, and the after-tax success rate, with federal and state tax, RMDs, IRMAA and Social Security taxation paid from the portfolio |
compare_return_assumptions | The same plan under each published capital market assumption set |
list_return_assumption_sources | Which assumption sets the engine carries, and what each publisher provides |
explain_methodology | What the engine models and what it deliberately leaves out |
estimate_2026_tax_return | A 2026 federal and state return estimate: the refund or amount owed, the main Form 1040 lines, the state tax, every default assumed and what is not modelled. Needs a QuantCalc PRO or Tax Season Pass licence key |
All six are read-only: they compute and return, and change nothing. Inputs are not kept after the run, except that a failed request is kept for two days to diagnose it.
What comes back with every number
A success rate is meaningless without its assumptions, and a summary will drop a caveat it was not handed explicitly. So each result states:
- the return model that actually ran — reported by the engine, not by the request;
- the number of paths and the real trial count behind the rate;
- the income it assumed, including when it assumed none;
- whether a correlation matrix had to be adjusted before running;
- a plain warning when a run is not precise enough to show a client;
- for the tax-aware tool: the spending convention (after tax — what the household keeps), every default assumed for an input that was not given, and the tax items the engine does not model.
Licence key
Optional. Without one, projections run at 2,000 paths (the 95% interval around a
success rate is about ±1.5 points) and the tax-aware tool returns the first
retirement year of its year-by-year schedule. A QuantCalc PRO key raises
that to 10,000 paths (about ±0.7) and returns the full schedule. The tax engine
itself — withdrawal order, Roth conversions, state tax — needs no key. The one
exception is estimate_2026_tax_return, which needs a PRO or Tax Season Pass
key (without one it returns nothing and points to the free web calculator at
https://quantcalc.app/2026-tax-return-calculator/). The same
key also unlocks the portfolio optimizer, glide paths, custom capital market
assumptions and multi-period planning in the
QuantCalc app. Those features are not tools in
this plugin.
Set it in the plugin's configuration. It is sent as a connection header and never passes through the conversation.
Privacy
The server receives only the inputs of each tool call, never the conversation. A random session identifier is used to count runs for each installation. A licence key is removed before anything is logged. Details are in section 6 of the privacy policy.
Not advice
QuantCalc is calculation software, not financial advice. Methodology and the source of every figure: https://quantcalc.app/methodology.html. Tax scope and its explicit exclusions: https://quantcalc.app/tax-methodology/.
Licence
MIT — see LICENSE. This repository contains the plugin manifest,
skills and documentation. The engine itself is a hosted service, published to the Official
MCP Registry as app.quantcalc/retirement-engine under a DNS-verified
namespace.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.