TL;DR

For retail AI-driven trading in 2026: Alpaca wins for solo operators (free paper + low-friction live, official MCP V2 with idempotency, options on a paid tier). Interactive Brokers wins for serious multi-asset-class operators (best asset coverage, lowest commissions at scale, most complete API surface — but the setup tax is real). Tradier fills the options-first retail niche. Robinhood and tastytrade APIs exist but have material gaps. Below: the head-to-head with idempotency + MCP coverage, the setup-tax reality, and who wins for which profile.

Criteria

For an AI-driven retail stack, broker selection hinges on:

  1. Idempotency on order submission — non-negotiable for any retry-capable agent.
  2. MCP server availability — official vs community; grade from /tools/finance-mcp-directory/.
  3. Asset-class coverage — equities minimum; options and futures desirable.
  4. Commission model — per-share, per-trade, zero-commission with PFOF, or subscription.
  5. API maturity — rate limits, documented error codes, paper/live parity.
  6. Paper-trading availability — for the entire pre-live validation period.
  7. Non-US availability — meaningful fraction of readership is EU-based.

The headline comparison

Broker MCP Idempotent Paper Commission Options Non-US
Alpaca Official V2 (grade A) Yes Free + full-featured $0 (PFOF) Yes (paid tier) US residents + some international
Interactive Brokers Community CLI (grade B) Yes Free, complete Tiered Yes Global
Tradier Community (grade C) No Yes $0 (PFOF) Yes — core offering US only
Robinhood None No No $0 (PFOF) Yes US only
tastytrade None Partial Yes $1/contract flat Yes — options-focused US only

Profile: solo retail AI operator, moderate volume

Alpaca Algo Trader Plus is the right default.

  • Official MCP V2 with 61 actions and idempotency support. Grade A.
  • Free paper trading with the same API surface as live (so prompt-blind research pipelines validate end-to-end before going live).
  • Options on the Algo Trader Plus tier ($99/mo); free tier for equities only (IEX feed).
  • PFOF model — commissions effectively zero but watch for effective spread costs.
  • SDK in Python is mature; retries, rate-limit backoff, error codes are well-documented.

Profile: serious multi-asset retail (equities + options + futures + international)

Interactive Brokers wins but levies a setup tax.

  • Community CLI MCP server (grade B, audit before production use).
  • Idempotency supported via client-supplied orderId.
  • Lowest commissions at scale (tiered; significantly below $0 PFOF effective cost above $10K notional).
  • Entire global market coverage — equities, options, futures, forex, bonds, funds.
  • Setup tax: requires TWS or IB Gateway running on the client machine. API is TCP to the gateway, not HTTPS to a vendor endpoint. Auth model is bearer token to the gateway process.

The gateway requirement is the single biggest friction for AI-driven setups. You either run the gateway 24/7 on your Mac/VPS or auto-start it via launchd. A fully headless gateway is technically possible but under-documented.

Profile: options-first retail

Tradier and tastytrade split the niche.

  • Tradier: community MCP (grade C, no idempotency), PFOF model, options are the core product. Good for options-first strategies if you can live with the community MCP limitations. API is reasonably documented.
  • tastytrade: $1 per options contract commission (transparent, no PFOF games). API is young — has improved dramatically in 2025–2026 but still missing some workflows. No official MCP; community servers not graded here (insufficient material).

For options strategies that require precise entry/exit timing and quotation integrity, tastytrade's transparent pricing is worth the commission premium over PFOF-based brokers.

Profile: EU-based retail trader

Interactive Brokers is the default (global access). Alpaca has been expanding international access but verify eligibility for your specific country. Tradier, Robinhood, and tastytrade are US-only.

Non-US retail traders also face meaningful regulatory differences in data reporting and tax treatment; see your local equivalent of Form 8949 / Anlage KAP.

MCP + broker alignment

The intersection of broker choice and MCP readiness narrows the field:

  • Grade-A MCP + broker: Alpaca (the only combination).
  • Grade-B MCP + broker: IBKR via community CLI MCP, for users willing to audit.
  • Grade-C MCP + broker: Tradier — workable for research, execution risky without manually patching idempotency.
  • No MCP: Robinhood, tastytrade — workable via direct SDK but no LLM-agent ergonomics.

For an AI-driven operator who wants to keep the LLM in the decision loop via MCP, Alpaca is materially ahead in 2026. Everyone else is either pre-MCP or community MCP only.

What to verify before committing

  • Paper/live parity. Does the paper-trading API surface match live exactly? For Alpaca and IBKR: yes. For Tradier: mostly.
  • Rate-limit behavior. What's the enforced rate? How are violations surfaced? Agents in retry loops can chew through rate limits invisibly.
  • Idempotency semantics. What happens if the same idempotency key is submitted twice within N seconds? Within hours? Within days? Some brokers dedupe within the session only; others dedupe permanently.
  • Commission + fee visibility. Does the API report the fee along with the fill? For precise backtest-to-live reconciliation, you need this.

Connects to

Sources