Alpaca vs IBKR — Migration Analysis for AI Trading Platform

Context

The current system is an Alpaca paper trading testbed running 24/7 on a dedicated Mac, with four automated bots and a Flask dashboard:

The goal evaluated here: migrate from Alpaca to Interactive Brokers (IBKR) to build a more sustainable, multi-strategy AI trading platform — one that can support different strategy types including stable income strategies, higher-frequency approaches, and AI-driven multi-source analysis. The account holder is HK-based and already has an IBKR account.

Fees Comparison

Execution Costs

For small share counts (50–100 shares), IBKR’s $1 minimum bites on every fill. For options, the gap is significant: a wheel strategy selling 1 contract/trade pays $0 on Alpaca vs $1.65 on IBKR — that’s ~$60/month at 1 trade/day before counting commissions on the buy-to-close.

Market Data

This is the hidden cost that flips the comparison for data-intensive strategies. A bot doing real-time AI analysis needs true NBBO, not IEX. IBKR’s data is both cheaper and higher quality.

Other Fees

Net Cost Verdict for Current Bots

For the existing strategy mix (US stocks + TSLA options), Alpaca is cheaper on execution. Zero commissions on options outweigh IBKR’s data cost savings. This changes if you move to strategies requiring real NBBO data or multi-asset instruments.

Asset Availability

Alpaca is a US equities/options sandbox. For HK residents, crypto live trading, futures, forex, and international stocks are all unavailable. Bonds exist but are limited to market orders on US government securities only — no secondary market, no corporates, no munis.

Beyond Asset Availability — The Real “Blocking” Factors

Even staying within US stocks, ETFs, and options, Alpaca has ceilings that affect strategy ambition:

1. Market Data Quality

Alpaca’s free tier is IEX — a single exchange covering roughly 15% of US equity volume. It is not the National Best Bid and Offer (NBBO), the consolidated quote from all exchanges. For AI systems doing real-time price analysis, price signals are incomplete. Strategies that depend on spread, momentum, or microstructure data will be working with degraded inputs.

IBKR provides true NBBO, Level 2 order book depth, tick-by-tick data, and historical tick data for backtesting — for ~$14.50/month total.

2. Order Type Sophistication

For strategies that need to minimise market impact (larger positions) or manage risk with bracket/OCA structures, Alpaca’s order type set is a hard ceiling.

3. API Rate Limits

Alpaca’s default is 200 requests/minute (~3.3/sec). For bots scanning many tickers, streaming multiple symbols, or running at higher frequency, this becomes a bottleneck. IBKR’s TWS API allows 50 messages/second (3,000/min) — 15× higher throughput.

4. Execution Quality (PFOF)

Alpaca uses payment for order flow — broker-dealer market makers pay Alpaca for the right to execute your orders. Those market makers profit from the bid-ask spread. For small retail trades, the impact is minimal. For a bot doing volume, execution quality erodes real P&L invisibly. IBKR Pro routes to the best venue across exchanges (SmartRouting) without PFOF.

5. Platform Longevity

Alpaca is a VC-backed startup (Series B). IBKR is a publicly traded company ($20B+ market cap) that has operated institutional trading infrastructure since 1978. For a platform you intend to invest serious development time into, broker durability matters.

“Blocking” — Decision Matrix by Strategy Type

The “blocking” is not just asset availability — it’s market data quality and order sophistication capping what AI strategies can actually do, even within US equities.

IBKR MCP Server Landscape

A survey of 13+ MCP (Model Context Protocol) servers for IBKR was conducted to assess AI integration options. All are unofficial and alpha-quality.

Summary Table

Best Option: YoungMoneyInvestments/ibkr-mcp

pip install ibkr-mcp

Capabilities: cancel_order, place_trailing_stop, get_option_chain (Greeks), bracket orders, OCA, TWAP/VWAP algo orders, market scanners, VaR calculation.

Confirmed gaps (small PRs to fix):

• No orderRef / client_order_id equivalent — 4-line addition per order tool
• No GTC time-in-force — add tif param to place_order and place_trailing_stop

Unattended auth: Pair with ibeam (812 stars, actively maintained) — Docker container that automates IB Gateway headless login including 2FA. Standard solution in the IBKR algo community.

Critical Insight: MCP ≠ Right Tool for Cron Bots

All current bots use this pattern:

cron → bash script → alpaca CLI → Alpaca REST API

There is no ibkr CLI equivalent. MCP servers are designed to be invoked by an AI agent as tools, not called from a shell script. For systematic cron-based bot execution, the correct tool is ib_insync — the mature Python library for direct TWS API access.

Recommended Architecture (If Migrating)

Strategy bots (Python scripts)
    └── ib_insync (TWS API, direct socket)
            └── IB Gateway (headless Docker via ibeam)
                    └── IBKR

Claude (AI decision layer — 3-phase pattern, unchanged)
    └── Phase 1: Shell/Python gathers market + political data
    └── Phase 2: claude -p receives structured prompt → returns JSON decision
    └── Phase 3: Python parses JSON → places orders via ib_insync

Dashboard (Flask — keep as-is)
    └── Swap run_alpaca() calls → ib_insync queries
    └── Attribution: use order.orderRef field (IBKR's equivalent of client_order_id)

What Can Be Reused from the Alpaca Testbed

Migration complexity estimate: 1–2 weeks for the bot execution layer rewrite. Dashboard and AI logic are portable as-is.

Verdict

The Alpaca testbed has served its purpose well. For a team building toward a production-grade, multi-strategy AI trading platform, IBKR is the right foundation — not because current strategies demand it, but because every ambitious direction you’ll want to explore will require it.

Open Questions for Team Discussion

1. Asset scope: Is the goal to stay within US equities/options, or expand to futures, forex, crypto, international markets?
2. Frequency: What is the target strategy frequency? Minutes = Alpaca fine; Seconds = IBKR needed.
3. Options commission priority: Is $0/contract a strategic priority for the wheel strategy? ($0.65/contract on IBKR is real cost at volume.)
4. Data quality: Does AI analysis require true NBBO, or is IEX sufficient for current signal generation?
5. Live trading timeline: When does paper trading graduate to live? IBKR’s live account infrastructure is more battle-tested.
6. Team ops: Can the team manage a persistent IB Gateway Docker container? Alpaca requires no local process.

Created: 2026-05-03