Claude for Financial Services: Agents & MCP (2026)

Claude for Financial Services is Anthropic's own open-source reference repository of agents, skills, and data connectors for workflows that come up constantly in investment banking, equity research, private equity, and wealth management. Everything in it ships from one source and runs two ways: as a Claude Cowork plugin, or as a headless deployment through the Claude Managed Agents API.

It's built by Anthropic itself (GitHub organization anthropics). As of this writing (September 28, 2026) the repository has roughly 37,988 stars and 5,478 forks, and it ranked #1 on GitHub's weekly trending list with +2,606 stars in a single week. It's licensed under the Apache License 2.0. This article covers the full lineup of agents, skills, and connectors, installation, basic usage, and what small businesses and accounting firms can learn from it, based only on the official README and GitHub repository.

The README is explicit about scope: "Nothing in this repository constitutes investment, legal, tax, or accounting advice. These agents draft analyst work product — models, memos, research notes, reconciliations — for review by a qualified professional. They do not make investment recommendations, execute transactions, bind risk, post to a ledger, or approve onboarding; every output is staged for human sign-off." Anyone using this is responsible for verifying the outputs and for complying with the laws and regulations that apply to their firm.

Each agent is named for the workflow it runs and ships as a self-contained plugin that bundles the skills it needs. The intended usage is to install the agents that match your work, then tune the prompts, skills, and connectors to how your firm actually operates.

You don't have to install a full agent — the underlying skills, slash commands, and data connectors are also available as vertical plugins on their own. The core financial-analysis plugin carries all 11 data connectors and the shared modeling skills; every other vertical plugin builds on top of it.

All connectors are centralized in the core financial-analysis plugin and shared across the rest of the repo. Each is exposed as an MCP server with a public URL, but most require a subscription or API key from the provider.

The repository itself is free and open source under the Apache License 2.0. But most MCP connectors require a separate subscription or API key from their provider, and Claude, Claude Cowork, and the Claude Managed Agents API are all governed by their own agreements with Anthropic.

In Cowork, either paste the repo URL in Settings or zip a directory under plugins/ and upload it. In Claude Code, add the marketplace first, then install plugins one by one.

To run agents headlessly as Managed Agents, set your API key and run the deploy script. scripts/deploy-managed-agent.sh resolves file references, uploads skills, creates leaf-worker subagents, and POSTs the orchestrator to /v1/agents.

- Install the core plugin: start with financial-analysis to get the shared modeling skills and every connector
- Pick the agents you need: install only the ones that match your work, such as Pitch Agent or GL Reconciler
- Point connectors at your own data: edit .mcp.json to connect to the MCP servers you actually have contracts with, such as Daloopa or FactSet
- Run it via slash commands: once installed, commands like /comps, /dcf, /earnings, and /ic-memo are available in your session
- Have a human review the output: agents only produce drafts. Final numbers and language should be verified by a qualified professional before use

For firms running Claude inside Excel, PowerPoint, Word, and Outlook, there's a separate Claude Code plugin called claude-for-msft-365-install. It's admin tooling that configures the Microsoft 365 add-in against your own cloud — Vertex AI, Bedrock, or an internal LLM gateway — rather than Anthropic's API, installed via /claude-for-msft-365-install:setup. Once the add-in is provisioned in a tenant, the agents and skills covered here run inside it. For more on Anthropic's broader agent ecosystem, see the Claude Code Complete Guide 2026 and the Claude Code MCP Integration Guide.

Claude for Financial Services is built for large financial institutions, but its design philosophy — drafts made for human review, not autonomous execution — and how it structures individual skills carry over directly to small-business IT and accounting/tax practices.

- The GL Reconciler / Month-End Closer pattern: finding breaks, tracing root cause, and routing for sign-off is a pattern that applies just as well to monthly closes and expense-report checks at a small business. The design deliberately doesn't automate everything — it hands a human a list of discrepancies and draft explanations
- The Statement Auditor pattern: putting an explicit audit step before distribution reduces the chance of a spreadsheet error slipping through. The same "AI prepares, a human does the final check" division of labor works for document review in accounting and legal practices
- The Skill & Command Reference structure: the README's pattern of defining a skill plus a slash command per workflow is a reusable template for turning your own back-office processes into Claude skills
- The MCP connector pattern: connecting your accounting software or cloud storage to Claude via MCP lets you build the same three-layer structure — connectors fetch data, skills do the judgment and drafting, a human makes the final call — for your own operations

Claude Cowork and Claude Code are general-purpose agent platforms that work across any industry; Claude for Financial Services is the finance-specific content you load on top of them. For Cowork basics, see the Claude Cowork Complete Guide; for enterprise security design, see the Claude Cowork Enterprise Deployment Guide. For the underlying MCP standard itself, see the MCP Roadmap 2026.

- As the README states, outputs never constitute investment, legal, tax, or accounting advice. Agents never make investment recommendations, execute trades, post to a ledger, or approve onboarding — everything is staged for human sign-off
- Most MCP connectors (Daloopa, FactSet, S&P Global, etc.) require a separate subscription or API key from the provider, so the free repository doesn't mean free data access
- Subagent delegation in Managed Agents (callable_agents) is a Research Preview feature; check each agent's README for security and handoff guidance before production use
- This is a template, not a finished product — the README itself frames it as something you tune to how your firm works, not something to run unmodified
- It's built for firms with finance-specific workflows — investment banking, PE, asset management, audit — that are already considering Claude Cowork or the Managed Agents API. Without those workflows, it's likely overkill

Claude for Financial Services is Anthropic's own reference implementation of the 10 workflows that come up constantly in investment banking, equity research, private equity, and wealth management, built on top of the general-purpose Claude Cowork and Claude Managed Agents API platforms. Ten named agents, nine vertical skill plugins, and twelve MCP data connectors all ship from one source, and every output is designed around human expert review.

Beyond direct use in finance, the underlying design pattern — let AI draft the work, surface discrepancies and risks, and leave the final call to a human — is a useful reference regardless of industry. A reasonable way to start is installing the core financial-analysis plugin plus one agent close to your own work, then pointing .mcp.json at your actual data sources.

- https://github.com/anthropics/financial-services
- https://claude.com/product/cowork
- https://docs.claude.com/en/api/managed-agents
- https://github.com/anthropics/financial-services/blob/main/LICENSE
- https://github.com/trending?since=weekly