Next-generation AI terminal: Claude Code connected to data on 17,000+ stocks and cryptocurrencies
An analyst under the pseudonym CyrilXBT demonstrated how, with a single command, Claude Code can be connected to live market data in under a minute—quotes for more than 17,000 stocks and cryptocurrencies, earnings reports, balance sheets, and cash flow statements. In his assessment, such functionality could replace the capabilities of a Bloomberg terminal.
This involves connecting an external financial data service via the MCP protocol. The setup consists of three steps and, according to the author, takes about 60 seconds.
What CyrilXBT offers
The first step is described in the analyst's post: inside Claude Code, you need to run a command to add the financial-datasets service via HTTP transport. The command specifies the MCP server address and registers it in the tool.
The second step is authorization. The user types /mcp inside Claude Code and completes the OAuth procedure in the browser. The connection status can be checked with a separate command that lists the linked services.
The third step is natural language queries. As examples, CyrilXBT cites questions about Apple's current price-to-earnings ratio and market capitalization, Tesla's quarterly reports for the last four periods, and the change in Bitcoin's price over the past year.
After setup, Claude Code, according to the author's description, directly accesses data on 17,000+ stocks, earnings reports, balance sheets, income statements, cash flow statements, and cryptocurrency quotes. Previously, this required a Bloomberg terminal, a complex financial data API, or hours of manual searching across various sources.
CyrilXBT separately highlights the cost of the alternative. In his estimation, analysts pay about $24,000 per year for a Bloomberg terminal, and the emergence of such a combination is unlikely to please these professionals.
Who benefits from this
According to CyrilXBT, a sustainable advantage will go to those who learn to combine Claude Code's reasoning with access to live financial data. This refers to quants, analysts, and portfolio managers.
Such an advantage, as the author notes, compounds every day. He frames it as a research edge that grows with daily use of the combination.
The analyst emphasizes that none of the steps is difficult on its own. The value arises precisely from connecting an ordinary tool to a stream of up-to-date market data in a single interface.
At the same time, the idea itself relies on an open standard for connecting external sources. Claude Code acts here as a shell through which the user accesses a third-party data provider without manual information gathering.
A thousand agents overnight from a smartphone
CyrilXBT also published a breakdown of the workflow of Boris Cherny, one of the creators of Claude Code. According to the author's description, Cherny uses a smartphone as his primary interface instead of a laptop and maintains between five and ten active sessions simultaneously.
Each of these can spawn subagents—sometimes several hundred at once, sometimes several thousand overnight on larger tasks. Dozens of loops, according to the author, run continuously in the background, with some of them persisting on the server side even when the laptop is closed.
CyrilXBT highlights four elements without which such a scheme is dangerous. The first is launching and delegating tasks from a smartphone: a full-fledged interface for sending instructions is needed, not just receiving status notifications.
The second is isolated parallel execution. Each agent or cluster of subagents operates in its own context and workspace; otherwise, running many processes simultaneously produces chaos, not scale.
The third element, which in the author's view is most often overlooked, is verification without human involvement. The result of each agent is checked against something real: another agent, a test suite, or an explicit rule—not the agent's own report of successful completion.
The fourth is strict stop conditions and a spending cap. A process running for hours unattended needs an absolute limit on time, money, and volume that does not depend on whether a person notices a failure. It is precisely this, as CyrilXBT emphasizes, that turns overnight agent work from a risk into a manageable task.
My take: connecting Claude Code to live financial data via MCP is not just a convenient hack but a signal of a paradigm shift in analytics. Bloomberg-level terminals have monetized exclusive access to information for decades; now that barrier is crumbling before our eyes. The question is not whether AI agents will replace analysts, but how quickly those who master such combinations will pull ahead of those who continue to work the old way. The data market is becoming a commodity, and the key competency is the ability to ask the right questions and build reliable verification pipelines.