Model Context Protocol · Bring your own LLM

Start chatting with your portfolios

Point the AI assistant you already use — Claude, ChatGPT, Gemini, or a local model — at your model portfolios. Ask what's in the book, what changed, and how it compares to your own positions. You form your own judgment, and — if you choose — bridge to your own brokerage to prepare orders. We don't provide the LLM, and we're never in your order path.

Read your holdings Compare to your own book Track what changed Pull the research

Connect your assistant View plans

Bring your own LLM

We are the data and research provider — not an AI provider. You bring the assistant you already use and pay for. Your prompts stay between you and your chosen model provider.

Claude (Anthropic)

Web, Desktop, or Claude Code — add our server as a custom connector.

ChatGPT (OpenAI) & others

ChatGPT, Gemini, Cursor, Windsurf — any client that speaks MCP.

A local model

Run open-weights models privately (e.g. via Ollama or LM Studio) and point an MCP-capable client at our server.

How to connect your assistant

Once you connect, anything the assistant reads leaves our security environment and is governed by your LLM provider's terms — not ours. Only connect an assistant you trust with your portfolio data.

1. Sign in with your account
Any account can connect and read our public research. An active subscription then unlocks your portfolios' holdings, history and rebalance tools — the connector always inherits your entitlements.
2. Add our server as a custom connector
In your assistant, paste the server URL below and sign in when prompted — it's OAuth 2.1, so there is no API key to copy.
3. Approve access and start researching
Approve the read-only scopes and ask about your holdings. Command-line clients (Claude Code, Gemini CLI, Cursor) use an X-API-Key token from API Access instead of the sign-in.

Connector details

Server URL (Streamable HTTP)

https://financial-portfolios.ai/mcp

In Claude, ChatGPT or any assistant with remote MCP support, add this URL as a custom connector and sign in when prompted — OAuth 2.1, no key to paste.

Command-line clients — use a token

CLIs and editor config files can't run the sign-in, so they authenticate with a personal token in the X-API-Key header. Generate one in API Access (shown once) and drop it in place of <your-token>.

Claude Code

claude mcp add -t http fpai https://financial-portfolios.ai/mcp -H "X-API-Key: <your-token>"

Gemini CLI

gemini mcp add -t http fpai https://financial-portfolios.ai/mcp -H "X-API-Key: <your-token>"

Cursor · Windsurf · other MCP clients (mcp.json)

{
  "mcpServers": {
    "fpai": {
      "url": "https://financial-portfolios.ai/mcp",
      "headers": { "X-API-Key": "<your-token>" }
    }
  }
}

Provider setup guides

Claude ChatGPT Gemini CLI Perplexity Cursor

These link to each provider's own documentation for adding a remote MCP connector.

What your assistant can read

list_portfolios

Every active product — marked as your subscribed (live) books or public (≈45-day delayed) ones.

get_portfolio_allocation

Target allocation (holdings + weights): the latest book if you're subscribed, otherwise the public delayed book.

get_portfolio_performance

How a model portfolio has performed against its benchmark, with risk statistics — total and annualized return, volatility, Sharpe, Sortino, maximum drawdown, beta and alpha. Hypothetical and simulated.

get_portfolio_metrics

Which holdings helped and hurt most, the names held most persistently, and win rates by name and by invested capital. Hypothetical and simulated.

get_portfolio_history

Past allocation snapshots over a chosen window — full for subscribers, delayed for public access.

get_portfolio_concentration

Concentration and sector mix — top-5 and top-10 weight, effective N, dominant sector. Derived from published weights, so available on the delayed book too.

get_portfolio_changes

What the model opened, closed, raised and cut over a period, plus turnover — what the model did, not returns.

rebalance_portfolio

Turn your own holdings into buy / sell / hold orders toward the model.

get_rebalance_template

A holdings CSV template for the rebalance input.

list_blog_posts / get_blog_post

Quantitative research articles and market commentary.

Interactive panels, right in the chat

In Claude and ChatGPT, the assistant renders these panels inside the conversation. In text-only clients the same data arrives as text — nothing is lost, only the interactivity.

Portfolio Explorer panel — Every holding, sorted how you like, without leaving the chat. — Portfolio Explorer
“Every holding, sorted how you like, without leaving the chat.”
Browse a model portfolio in the conversation: a sortable, filterable holdings table, sector weights, a concentration curve, what changed over a period, a side-by-side comparison of two or three portfolios, and a search for which portfolios hold a given stock.

Access & Connection panel — What your subscription shows you, and what it doesn't. — Access & Connection
“What your subscription shows you, and what it doesn't.”
Which portfolios the account can see, which are live and which are delayed by about 45 days, and exactly which figures a delayed book leaves out — rather than discovering blanks. Also shows how the connection is signed in.

Rebalance Studio panel — Your holdings, checked by you, before any arithmetic runs. — Rebalance Studio
“Your holdings, checked by you, before any arithmetic runs.”
Compare your own holdings against a model. If the assistant can read positions from a connected brokerage it fills them in; otherwise paste them from a spreadsheet. Either way they appear in an editable table for checking first, because a brokerage read can be stale, partial, or from the wrong account.

Research Reader panel — Our research, filtered to what you own. — Research Reader
“Our research, filtered to what you own.”
Browse, search and read the research in place, with posts flagged when they mention a ticker your portfolios actually hold.

Breadth Board panel — What the strategies agree on. — Breadth Board
“What the strategies agree on.”
What the portfolios have in common: the most widely held names, how much any two books overlap by name and separately by weight, and each one's dominant sector and concentration.

Where we sit in your stack

We publish the model portfolios. Market data, news and brokerage access come from services you already use. The assistant joins them — so “what’s the news on my biggest holding?” and “how far has my account drifted from the model?” become single questions instead of three tabs.

Financial Portfolios AIus

Model holdings, weights, changes, concentration, performance and research — what we publish.

Market data & news

Company identity, news, filings and fundamentals — from a market-data service you already use.

Brokerage / aggregator

Your own account positions — from your broker, or an aggregator you connect.

Nothing trades

The connector computes arithmetic and places no orders. Order tickets are proposed for your review; you confirm and execute. There is no auto-execute anywhere.

Delayed users are not second-class

Concentration, sector mix, effective N and breadth are derived from published weights, so they appear on the public delayed book too — even where the raw API returns nulls.

Try these prompts

Once connected, just ask in plain language — your assistant calls the right tool for you. Click any prompt to copy it, then paste into Claude, ChatGPT, Gemini, or your MCP client.

See your holdings

Track changes & history

Research & compare

Align your own holdings

Tip: if your assistant answers from general knowledge instead of your actual book, name the portfolio (e.g. "AQA Balanced") or say "using the Financial Portfolios connector" to nudge it to call the tool.

Works with

Claude (web, desktop & Code)ChatGPT (Custom GPT Actions)Google GeminiCursorWindsurfAny MCP client

Remote connectors use Streamable HTTP; local clients can also run over stdio. Copy-paste setup for each client lives in API Access.

Works across Claude, ChatGPT, Perplexity & Gemini

Our server is standard MCP + OAuth 2.1, so any assistant that supports remote MCP connectors can read your portfolios. Capabilities differ by platform:

Claude & ChatGPT — full support, and the strongest for the optional brokerage bridge (Interactive Brokers, Robinhood, Webull all offer official connectors).
Perplexity — MCP connectors on Pro / Max / Enterprise; great for research, but brokerage access there is read-only (no order placement).
Gemini — MCP support is maturing and is strongest for market-data connectors; retail brokerage execution lags the others today.

From research to your own order

Two separate connectors, both yours: our read-only research MCP, and — if you want to act — your own broker's MCP. You stay in control at every step.

1
Connect our MCP (read & research)
Your assistant reads the model portfolios — the latest book for products you subscribe to, and the public delayed book (≈45 days) for the rest — plus allocations, history, and research. Nothing beyond what your entitlements allow.
2
Research the holdings with your LLM
Ask your assistant to explain the book, compare it to your own positions, summarize what changed, or dig into a name. The analysis is yours to interrogate.
3
Form your own judgment
You decide what — if anything — to do. Our data is informational and model-level; it is never a recommendation to buy or sell.
4
Bring your own brokerage (optional)
Separately connect your broker's own MCP (Interactive Brokers, Robinhood, Alpaca, …). That connector is between you and your broker — we are not part of it.
5
Review & place every order yourself
Keep review-before-execute on. Verify each proposed order, then approve it in your broker. You place the trade; we never do.

Bring your own brokerage

To act on your judgment, connect your own broker to your own LLM through their MCP connector. Several brokers now ship official connectors. Details, capabilities, and cautions differ — pick a tab.

The broker-by-broker details below — including each broker's own risk warnings and capability notes — are shown in English so their meaning is not altered by translation.

These brokerage connectors are operated by the brokers or third parties — not by Financial Portfolios AI. We do not build, host, or endorse them, and we are never in your order path. Connecting one links your brokerage account to your LLM. Keep review-before-execute on, start in paper mode, and place every order yourself.

Interactive Brokers

Official broker connector

IBKR ships an official MCP connector (2026). You link your account through the platform's certified connector marketplace — no API keys or passwords are shared with the AI tool.

Endpoint: api.ibkr.com/v1/api/mcp-public (or the certified Claude / ChatGPT / Grok connector)
Assets: Global stocks, ETFs, options, futures, FX and bonds — per your IBKR account.
Authentication: Certified connector authorization in your IBKR account — not raw API keys/passwords handed to the LLM.
Reads: Positions and unrealized P&L, cash balances and buying power, real-time quotes and 30-day history, trade history, and open orders.
Can place orders?: Drafts trade instructions only (e.g. “buy 100 AAPL at a $200 limit”) and sends them to your IBKR account for review and approval. Per IBKR: the assistant never submits orders directly to the market.
Paper / dry-run: Yes — connect an IBKR paper-trading account to rehearse the whole flow with no real money.

Before you connect

IBKR states displayed symbols are illustrative only and not recommendations; IBKR is Member SIPC.
Approve every drafted order yourself, and confirm the connector is linked to the account you intend (paper vs live).

IBKR Claude connector IBKR AI integrations

Secure & scoped

Every request to our MCP is authenticated — by OAuth 2.1 for connectors, or an API-key header for command-line clients — and served under the same delayed / entitlement rules as the rest of the platform, so an assistant only ever sees what your subscription allows. Revoke access any time from API Access. We never receive your LLM prompts or your brokerage credentials.

Important disclosures for agentic & brokerage use

The important agentic & brokerage disclosures below are provided in English only.

Bring your own LLM — we don't provide one

We do not provide, host, resell, or endorse any AI model. Claude, ChatGPT, Gemini and local models are operated by their respective providers under their own terms. Whatever you type, and any data you choose to share with the assistant, goes to the provider you pick — not to us.

Not investment advice

Everything our MCP returns is informational and educational, simulated or delayed, and describes a model — not a recommendation to buy or sell any security for you. No suitability assessment has been performed. You are solely responsible for your own decisions.

AI assistants make mistakes

LLMs can misread data, hallucinate figures, and generate incorrect or unsafe orders. Independently verify every number and every proposed trade before acting. Never place an order you do not fully understand.

We are never in your order path

Any brokerage connection is strictly between you and your broker (via that broker's or a third party's MCP). We do not receive, route, confirm, or execute orders, and we never touch your brokerage credentials.

Brokerage connectors are not ours

The Interactive Brokers, Robinhood, Alpaca and other connectors described here are operated by those brokers or third parties. We neither build, host, audit, nor endorse them. Review each connector's terms, permissions, and security before use.

Keep a human in the loop — prefer propose → confirm

Our marketplace tooling prepares order tickets for your review; it is not designed to trade automatically. Some broker connectors can execute without confirmation if you instruct them to — don't. Use paper/dry-run modes first and keep review-before-execute enabled.

Trading involves significant risk

Agentic and self-directed trading can lose money, including your entire investment. Past and simulated performance does not indicate future results. Only fund an agentic or brokerage account with money you can afford to lose.

We are a publisher, not your broker or adviser

Financial Portfolios AI publishes model portfolios and research. We are not a broker-dealer, investment adviser, or custodian; we do not manage money, place trades, or personalize recommendations.

This page supplements — and does not replace — our full disclosures. By connecting any assistant or brokerage you accept these terms and act on your own responsibility.