Anthropic open-sources Wall Street

The interesting part here is not that Anthropic shipped “AI for finance.” Everyone is doing that. The real news is that Anthropic published a repo that shows how enterprise agents are actually wired for a hard, regulated industry — and developers immediately treated it like a playbook, not a promo page. GitHub now shows the `financial-services` repo at roughly 16.7k stars, which is a huge reaction for something this specific. ### What did Anthropic actually open-source? It wasn’t a model. It was the workflow layer — the prompts, agent structure, task breakdowns, and integration pattern for finance jobs like pitch-book creation, market research, earnings review, valuation review, KYC screening, month-end close, and statement auditing. The repo sits in Anthropic’s public GitHub org under an Apache 2.0 license, which makes the intent pretty clear: copy this, adapt it, and build on top of it. (github.com) ### Why are people calling it “Wall Street”? Because the agent names map directly onto real analyst and operations work. The YouTube coverage that pushed this story framed it as Anthropic “open-sourcing Wall Street,” which is a little dramatic, but you can see why people say it. The repo reorganized into 10 end-to-end agents — including Pitch Agent, Market Researcher, Model Builder, Meeting Prep, GL Reconciler, and KYC Screener — basically the software version of an investment bank org chart. (github.com) ### What makes the repo more than a demo? The connectors. Anthropic didn’t just mock up fake tools. The repo is described as wired to 11 licensed MCP data connectors, including FactSet, S&P Global, Morningstar, Moody’s, PitchBook, LSEG, Daloopa, Aiera, Chronograph, MT Newswires, and Egnyte. That means the open-source part is the orchestration and skill layer, while the expensive proprietary data still sits behind licensed access. In plain English — Anthropic published the recipe, not the pantry. (youtube.com) ### Why does MCP keep showing up in this story? Because MCP is the part that lets an AI app talk to tools and data sources in a standard way. Anthropic introduced MCP in November 2024 as an open standard for connecting assistants to business systems, repos, and dev environments. The official docs now describe it as a common way to hook LLM apps into external data and tools, and the registry shows a fast-growing server ecosystem. So when this finance repo ships with MCP plumbing, developers read that as reusable infrastructure, not just one-off finance code. (youtube.com) ### Where does A2A fit in? A2A solves a different problem. MCP is mostly about tool use and context. A2A is about one agent talking to another agent. Google introduced Agent2Agent in April 2025, then donated it to the Linux Foundation that June, and Google later said the ecosystem had grown past 150 supporting organizations. That is why people keep pairing the two ideas: MCP helps an agent use systems; A2A helps multiple agents coordinate. (anthropic.com) ### So what changed this week? The shift is from protocol talk to visible implementation. For months, developers have heard that agents need standards. Now one of the biggest AI labs has published a concrete, high-status example in a domain where bad automation gets expensive fast. Finance is useful here because the work is structured, document-heavy, and tool-dependent — exactly where standards either prove themselves or fall apart. (developers.googleblog.com) ### Is this really about finance? Partly. But mostly it’s about legitimacy. Open-source agent projects often look clever and fragile. Anthropic’s repo looks like enterprise middleware with domain opinions attached. That gives builders a template for other verticals — legal, healthcare ops, insurance, procurement — where the winning move may be less “build a better chatbot” and more “standardize the handoffs between models, tools, and specialist agents.” (youtube.com) ### Bottom line? Anthropic didn’t open-source Wall Street’s data moat. It open-sourced a chunk of Wall Street’s operating grammar. And that matters because the agent race is starting to look less like a model race and more like a standards race — whoever defines the plumbing gets pulled into everything built on top. (youtube.com) (github.com)