Are model weights allocation advice?

No. They are a deterministic historical scenario under stated constraints and must not be treated as personalized targets or orders.

Can AI predict stock prices with this MCP tool?

No. AI cannot predict stock prices with certainty, and this MCP tool does not promise future returns. It gives an agent bounded data or validation results that still require independent review, risk controls, and source checking.

Agent workflows

Portfolio Optimization MCP Tool for AI Agents

Q: Are portfolio inputs or model weights cached?

No. The final workflow bypasses Agent result-cache keys, entries, single-flight, persistence, and portfolio-derived logs or public errors.

optimize-portfolio builds a deterministic historical model scenario for 2–15 active US stock, ADR, or ETF symbols. It supports minimum volatility, risk parity, and maximum Sharpe objectives, adjusted common-date returns, explicit max-weight and one-way turnover constraints, and a chronological 70/30 development/holdout comparison.

Connect Pineify MCP View all MCP tools

MCP tool name

optimize-portfolio

Example prompt

Use Pineify optimize-portfolio for 50% SPY and 50% QQQ with minimum volatility, 75% max weight, and 30% turnover.

Direct answer

What optimize-portfolio does

Current positions must sum to 100% within 0.01. Up to five zero-start candidate symbols may be added without exceeding fifteen total assets.

The final portfolio request and result bypass Agent result-cache keys, entries, single-flight, persistence, logs, analytics, and public error details. Provider-level market-data caches may still operate independently.

Contract

What the agent sends and receives

Inputs

`positions`: 2–15 unique symbols with positive current weights totaling 100%.
`candidateSymbols`: up to five unique zero-start assets; total assets remain at most fifteen.
`objective`, 1y or 3y `lookback`, `maxWeightPercent`, `maxTurnoverPercent`, and optional `detail`.

Structured output

`portfolio-optimization-v1` structured content with redacted query counts rather than original input.
Included/excluded coverage, current/model weights, deltas summing to zero, turnover, and development objective.
Same-window holdout comparison, assumptions, warnings, and explicit final-cache bypass.

Capabilities

Where this MCP tool fits

Solve three fixed objectives

Uses deterministic long-only minimum-volatility, risk-parity, or mean-shrunk maximum-Sharpe methodology.

Enforce explicit constraints

Keeps weights fully invested while respecting max weight, one-way turnover, finite, and convergence checks.

Compare holdout outcomes

Reports current and model return, volatility, Sharpe, and drawdown on the same untouched holdout sample.

Agent workflow

A bounded call from question to review

1
Define a complete current portfolio
Send unique exact symbols and current weights that sum to 100%.
2
Choose objective and constraints
Set one fixed objective, lookback, max weight, and turnover cap.
3
Compare rather than obey
Review exclusions, development objective, and both holdout outcomes before making any independent decision.

Prompt examples

Questions an agent can route to this tool

"Use Pineify optimize-portfolio for 50% SPY and 50% QQQ with minimum volatility, 75% max weight, and 30% turnover."

"Compare a risk-parity historical scenario for AAPL, MSFT, and NVDA over 3y without treating model weights as advice."

Operator notes

How I review the result

I keep excluded assets and weaker model holdout performance visible.

I do not convert model weights into shares, orders, targets, or personalized suitability advice.

Boundaries to keep in the prompt

The model is long-only, fully invested, historical, and limited to three fixed objectives.
Insufficient common observations, infeasible constraints, or non-convergence return bounded errors rather than approximate targets.
Taxes, transaction costs, account circumstances, and suitability are not modeled; the result is not investment advice.

Pineify MCP is an information and code-validation tool, not investment advice. It does not promise returns, place trades, or modify a portfolio. Review timestamps, source fields, code diagnostics, and risk assumptions before acting.

MCP tool family