Alpaca's Claude Connectors Simulate Trades but Cannot Execute Them

Alpaca launched two connectors for Anthropic's Claude on November 12, 2026, and neither one can place a live trade. The Paper Trading connector can submit simulated orders in stocks, ETFs and options; the Live connector can read balances, positions and market data from a real account. Live execution remains confined to Alpaca's Trading API or its Dashboard, a design choice that defines what an AI assistant may do in a brokerage workflow and what it must not.

The launch matters because it draws a precise boundary in agentic finance (AI systems that take actions on a user's behalf through external tools). Brokerage APIs are increasingly exposed to conversational interfaces, and Alpaca's implementation shows a vendor deliberately splitting informational access from execution. Alpaca already operates an MCP server (Model Context Protocol, a standard for connecting AI assistants to external tools), a command-line interface and other AI connectors; the two Claude integrations extend that stack.

What Can Each Connector Actually Do?

The Paper Trading connector links a Claude conversation to an Alpaca paper account through the company's Trading API and Market Data API. It supports market analysis, account monitoring and simulated trades in stocks, ETFs and options. Its listing in Anthropic's Claude directory also includes crypto. The Live connector connects to a live Alpaca account for portfolio monitoring and market research, and it can create or change watchlists.

Both connectors can report balances, positions and account activity and analyze market trends. The permission split is explicit: only Paper Trading can submit orders, and those orders use simulated funds. Neither connector can submit a live order.

The Live connector's directory listing offers prompts to summarize balances, buying power, open positions and portfolio concentration, or to identify which positions drove equity changes over the past 30 days. It can also retrieve real-time quotes and historical market data. Buying power is the amount an account can use for orders under its current limits.

How Much Should Users Trust the Responses?

Both directory pages label the integrations as community connectors. Their notices state that the connectors receive automated checks but are not verified by Anthropic, which does not control how their tools work or change. That classification places responsibility for behavior and accuracy outside the assistant's developer.

Alpaca's own guidance is blunt about verification. Connecting either service requires an Alpaca sign-in and authorization flow, and users are instructed to review the requested permissions and confirm the intended account before authorizing access. For paper trading, Claude's account response should be compared with the paper account in Alpaca's dashboard. Users with multiple paper accounts must confirm which one is connected.

The company also sets a hard rule for discrepancies: any account or balance mismatch should be resolved before proceeding, and an order should not be used to test an unresolved mismatch. Alpaca instructs users to review AI-generated order details before submission and independently verify order statuses, balances and positions, noting that responses can contain errors, omissions or inaccuracies.

One operational trap deserves emphasis. An order acknowledgment does not establish a fill. The order may remain open, fill partially or be rejected, so the returned status and any rejection message need inspection rather than a conversational confirmation treated as proof that a position changed.

How Realistic Is the Simulation?

Paper trading uses simulated funds rather than real capital. Alpaca's paper-trading documentation describes a real-time environment that is free and available to all Alpaca users, with orders following the live-trading workflow except that they are not routed to a live exchange. Fills are simulated from real-time quotes.

The simulation has documented limits. It does not account for:

• market impact
• order-information leakage
• latency-related slippage
• queue position for non-marketable limit orders
• price improvement
• regulatory fees
• dividends

A completed paper order therefore demonstrates the simulated workflow. It does not predict a live fill or establish how a strategy would perform with real money.

What Determines Market-Data Access?

Market-data access is a separate check from account access and order execution. What Claude can retrieve depends on the authenticated account's subscription, permissions and enabled asset classes. A successful balance request does not establish access to every symbol, feed or asset class. Options data appears only for accounts with that asset class enabled, and reconnecting alone may not resolve a missing permission or unsupported tool.

The Market Data API documentation spells out the subscription tiers. The free Basic plan is the default for paper and live Trading API accounts, but its real-time coverage is limited to IEX (the Investors Exchange, a single US stock exchange) for equities and an indicative feed for options. Algo Trader Plus provides real-time equities coverage across all US stock exchanges and the OPRA options feed (the consolidated US options market data system).

The gap widens on historical data. Basic excludes the latest 15 minutes of historical equities and options data and permits 200 historical API calls per minute. Algo Trader Plus removes that time restriction and raises the limit to 10,000 calls per minute, a fiftyfold difference.

What Does This Mean for Traders and Developers?

For developers and traders bringing these workflows into Claude, the practical capability is defined by more than the conversational interface. Account permissions, simulation assumptions and data subscriptions determine which analyses and order tests the connectors can support. The tiers matter: an analysis run on IEX-only equities data with a 15-minute historical gap will differ from one with full US exchange coverage.

Alpaca remains the venue for live execution and independent verification. The connectors' arrival on November 12, 2026 marks a market where brokerage infrastructure is being conversationalized at the edges while execution, compliance and data rights stay with the licensed venue. Developers building on the stack will need to test against all three gates — account authorization, simulation assumptions and data subscription — before treating any Claude-generated analysis as actionable.