Financial Datasets Brings Sub-200-Millisecond Stock Data to AI Coding Assistants
Financial Datasets, a startup that says it is backed by Y Combinator, has launched a real-time stock API targeting developers who build AI agents, offering what it describes as sub-200-millisecond access to price data, fundamentals, and SEC filings. The New York-based company went live earlier this year, positioning itself as an alternative to legacy providers FactSet and S&P Global in a market that's growing crowded with rivals chasing the same AI-native opportunity.
The platform combines more than three decades of financial KPIs across some 27,000 U.S. stocks with a hosted Model Context Protocol server that connects directly to Claude, Cursor, and other AI coding assistants. It's a bet that developers building agentic tools will pay for speed and structured data rather than scrape public sources or wrestle with incumbent vendors' clunky interfaces.
Financial Datasets exposes real-time stock snapshots, historical open-high-low-close-volume data, standardized fundamentals, SEC filings with extracted line items, insider trades, institutional holdings, and earnings releases. The company says it achieves 99.99% precision and availability, figures it reports are verified against SEC EDGAR filings and press releases across a rotating 1,000-company sample each audit cycle.
"I think we have the best Earnings API on the market," founder Virat Singh wrote on LinkedIn in late July. "Structured earnings are ready in our API ~3 seconds after a company reports, at 99.99% precision."
Singh previously held engineering roles at Faire, Airbnb, Acorns, GoPro, and Sony, according to a third-party profile on getprog.ai. The company lists between two and 10 employees on LinkedIn, though it declined to disclose funding details beyond its Y Combinator backing.
The startup announced a partnership with Julius AI in late March, enabling Julius users to query live market and fundamental data through natural language prompts. Use cases include building comparables tables, running debt coverage analyses, and breaking down segment revenue. Financial Datasets also published a benchmark called EarningsBench in June, testing six large language models on SEC filing extraction. Anthropic's Opus 4.6 with Extended Thinking scored 99.3% accuracy across 60 earnings releases spanning a dozen sectors, the company reported. "Our accuracy bar at Financial Datasets is 100%," the company wrote in a blog post announcing the results.
Pricing starts at $20 for 1,000 API requests under a one-time credits plan, climbing to $200 monthly for 100,000 requests under a tier called "Build." Enterprise customers needing webhooks, zero data retention, and uptime service-level agreements face custom pricing. Premium endpoints consume credits at 4× or 8× multipliers depending on complexity.
The move puts Financial Datasets in direct competition with developer-focused providers including StockContext, Tidore, Intrinio, and Trellis, all of which have added MCP servers or OpenAPI catalogs for agent integrations. Larger incumbents are making parallel moves. FactSet announced what it called "the industry's first production-grade model context protocol server" in mid-December, while S&P Global launched Adaptive Retrieval over the summer.
"The use of AI in financial services is rapidly accelerating and evolving, from tightly controlled workflows to fully autonomous, multi-agent systems," Sally Moore, chief client officer and co-head of Market Intelligence at S&P Global, said in a press release at the time.
Financial Datasets has recently added 72 key financial ratios, expanded coverage to more than 17,700 global stocks, and posted SpaceX financials extracted from the company's S-1 filing, according to LinkedIn updates from the past month. Whether precision and speed alone can dislodge established vendors with decades of institutional relationships remains an open question, but the startup is betting that developers building the next generation of financial tools will choose APIs designed for AI over retrofitted legacy systems.
