Charles Schwab, the US brokerage with $13.33 trillion in client assets in the third quarter of 2026, announced a Schwab MCP server on 12 November 2026, letting clients connect their own AI platform to read account data and, on thinkorswim accounts, place trades with the client in the loop. The Schwab MCP server runs as a pilot for chosen clients on Anthropic’s Claude, with wider access promised but no dates or client counts. Fintechbits checked Robinhood’s help centre and found that Robinhood’s Trading MCP server accepts Claude, ChatGPT, Grok, Cursor and Perplexity, while Schwab’s pilot supports Claude alone. For a brokerage of that size, the Schwab MCP server is a late and cautious entry rather than a lead.
What the Schwab MCP server lets a client’s AI do
The Schwab MCP server is a connection layer. Model Context Protocol is an open standard that lets AI agents connect to external apps and services, and Schwab is publishing its own endpoint for clients to connect to. In Schwab’s announcement, a client authorises the connection and the AI can then pull market data and research, review how positions have performed, summarise a portfolio, set alerts and organise investment notes.
Accounts that are enabled for the thinkorswim platform get a further step. Their AI can prepare, submit and manage orders, with the client kept in the loop. Schwab does not define what that means in practice. The announcement is silent on whether each order needs an explicit click, how long an authorisation lasts, and which tools the server exposes. A disclosure adds that Anthropic has no affiliation with Schwab’s broker-dealer.
Schwab’s Managing Director and Head of Trading Services, James Kostulias, framed the package as a way to give clients “more choice in how they engage” with the firm. The same announcement adds extended-hours options on about 30 products and a thinkorswim Assistant, but the MCP server is the item that changes who sits between a client and an order ticket.
Schwab MCP server against Robinhood and Interactive Brokers
Robinhood moved first of the three. Its 27 May 2026 announcement opened Agentic Trading to outside agents through a dedicated account separate from the rest of the client’s portfolio, with equities only at launch. That is 169 days before the Schwab MCP server pilot. Robinhood’s own disclosures say orders may be placed by an AI agent without the customer’s input on each transaction, which is a different promise from Schwab’s in-the-loop wording.
Interactive Brokers sits closer to Schwab. Fintechbits reported that the broker connected client accounts to Claude on 1 June 2026, added ChatGPT and Grok about three weeks later and opened to almost any MCP-capable tool on 28 July 2026, always with a human converting the AI draft into an order, as Fintechbits covered in July. Schwab is therefore matching the human-approved tier that Interactive Brokers already occupies, a notch behind Robinhood’s autonomous design, and doing it with a single supported AI platform.
Where Charley fits beside the Schwab MCP server
Schwab announced its in-house assistant Charley on 30 September 2026, with a phased rollout from October. As Fintechbits reported at the time, the actions Charley can complete are housekeeping such as watchlists, alerts, beneficiaries and contact details, while transfers, payments and trading appear only as topics for guidance. The Schwab MCP server reverses that boundary. Order entry is possible, but through a model Schwab does not build, host or control.
That split is the main design choice. Schwab keeps its own generative assistant away from money and orders, and lets clients take the model risk on themselves through an outside AI. It is a defensible way to learn from a small pilot before committing the Schwab brand to agent-driven trading.
Fintechbits analysis
Why the Schwab MCP server matters even as a narrow pilot
The Schwab MCP server matters because a brokerage holding $13.33 trillion in client assets has opened its accounts to outside AI, and its answer sets a conservative template. Schwab reports $13.33 trillion of client assets and an average of 10.7 million trades a day in Q3 2026. If even a small share of those clients route research and order staging through a chat window, the number of orders that begin as a sentence rather than a ticket rises quickly.
The pilot is small enough to count as a test. One supported AI platform, a select group of clients and no published tool list leave the Schwab MCP server short of a launch, and Fintechbits would not treat it as one until Schwab names a second platform and a client count. For rivals, the message is that outside AI access is now expected. For clients, the open question is liability: Schwab’s announcement does not say who bears the loss when a model misreads an instruction, while Robinhood’s help centre says the customer is responsible for trades an agent places.
Frequently asked questions
What is the Schwab MCP server?
The Schwab MCP server is Charles Schwab’s own Model Context Protocol endpoint, announced on 12 November 2026. It lets a client’s AI platform access that client’s Schwab account data after the client authorises the connection.
Who can use the Schwab MCP server?
Only a select group of Schwab clients who use Anthropic’s Claude can use the Schwab MCP server during the pilot. Schwab says access and supported AI platforms will expand, without giving dates, and order functions require a thinkorswim-enabled account.
Can the Schwab MCP server place trades?
Yes, for thinkorswim-enabled accounts, the Schwab MCP server can prepare, submit and manage orders through the client’s AI platform. Schwab says the client stays in the loop, but it has not said whether every order needs a separate approval.
What to Watch Next
The first marker for the Schwab MCP server is a second supported AI platform, since Schwab has promised more choices and the Robinhood help centre already lists several. The second is documentation: a public tool list, permission scopes and revocation steps would show how much control a client really keeps. The third is scale, meaning the first disclosed count of pilot clients or connected accounts. Last is whether the in-the-loop rule loosens, because Robinhood has already shown that autonomous orders are possible in a ring-fenced account.



