Anthropic Finance Agents: 10 Templates, New MCP Connectors & What They Cover (2026)

What Anthropic Launched in May 2026

In May 2026, Anthropic announced a dedicated push into financial services: ten finance agent templates, a new suite of official MCP connector partnerships with major financial data providers, and Microsoft 365 integration (Claude add-ins for Excel, PowerPoint, Word, and Outlook). The announcement named Citadel, BNY, Carlyle, Mizuho, Travelers, Walleye Capital, FIS, and Hg as customers already using Claude in financial workflows.

This is the clearest signal yet that Anthropic views financial services as a primary vertical — and it has direct implications for every finance team evaluating Claude for professional use. This article breaks down what each template covers, what the new connectors provide, and — critically — where the coverage gaps remain.

The 10 Finance Agent Templates

Research & Coverage Agents

Pitch Builder

Builds pitch decks and investment books from company data. Ingests financials, market data from connected providers (FactSet, S&P Capital IQ), and news to produce structured pitch narratives. Best suited for sell-side coverage initiation, buy-side investment committee memos, and M&A marketing materials. The agent coordinates between data retrieval, model outputs, and formatting subagents to produce a structured deliverable.

Earnings Reviewer

Analyzes earnings calls and financial results. Takes a transcript or filing as input, extracts key metrics (revenue, EBITDA, EPS vs. consensus), identifies management commentary themes, and flags guidance changes. Links to prior quarters for YoY comparison. The output is a structured earnings summary in analyst-report format — faster than reading the 100-page 10-Q, with key variances surfaced automatically.

Model Builder

Builds financial models from user inputs. Accepts company financials, growth assumptions, and margin drivers to produce three-statement forecasts, DCF valuations, or LBO structures. Connects to market data providers for public comps. This is the most technically demanding of the five research templates — model quality depends heavily on the assumptions the user provides and whether the template has been configured for the specific asset class.

Market Researcher

Synthesizes industry, competitive, and macro data for a sector or company. Pulls from connected data providers (IBISWorld for industry sizing, Guidepoint/Third Bridge for expert insights) and synthesizes into structured briefings. Replaces the analyst task of pulling from six different databases and writing a coherent narrative.

Meeting Preparer

Briefs analysts and bankers for client meetings. Takes a company or counterparty name, pulls recent financials, news, deal history, and relationship context (from integrated CRM if configured), and produces a pre-meeting briefing in a consistent format. The Microsoft 365 integration means this can push directly to an Outlook calendar invite or a Word document.

Finance & Operations Agents

Valuation Reviewer

Reviews valuation models for consistency, methodology, and defensibility. Checks DCF assumptions (WACC, terminal growth rate, projection period), comps selection and spread, and LBO structure. This is an audit function — it doesn't build the model, it reviews one you have and flags where the assumptions are aggressive, where the comparables diverge, and where the methodology departs from market practice.

General Ledger Reconciler

Reconciles general ledger accounts. Matches transactions, identifies posting differences, flags items requiring investigation. For firms already using Claude Enterprise with official QuickBooks or Xero MCP connectors, this agent can pull live GL data for the reconciliation rather than working from exports.

Month-End Closer

Orchestrates the monthly close checklist — journal entries, accruals, reconciliations, variance explanations, management reporting package. The agent sequences tasks, assigns them, and tracks completion. Equivalent to a digitized close checklist but with AI-generated commentary for each step.

Statement Auditor

Reviews financial statements for consistency, disclosure compliance, and anomalies. Checks cross-statement ties (net income flows through equity, cash flow reconciles), flags unusual line items, and checks disclosures against applicable standards (GAAP or IFRS). Most useful as a first-pass audit efficiency tool before the formal review.

KYC Screener

Screens customers and counterparties for KYC/AML compliance. Connects to D&B, Verisk, or OFAC data to check business identity, beneficial ownership, sanctions lists, and adverse media. Produces a structured KYC file for compliance officer review. Replaces the manual multi-source screening process that typically takes 30–60 minutes per entity.

New Official MCP Connectors

Alongside the templates, Anthropic announced a set of official MCP connector partnerships that provide structured data access for the agents. These connectors are available in the Claude financial services marketplace and integrate with the templates out of the box.

Existing Anthropic financial data partners include FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Chronograph, LSEG (formerly Refinitiv), and Daloopa. The full connector ecosystem now covers most major institutional financial data providers — if your firm already subscribes to these platforms, connecting to Claude via official MCP is the path of least resistance.

Availability and Requirements

Three important constraints that affect whether these templates apply to your team:

1. Paid Claude plan required — Plugins and templates work on Claude Pro and above. Managed Agents (the most autonomous variant) require Claude Team or Enterprise and are currently in public beta.
2. Enterprise data subscriptions required — The MCP connectors for FactSet, S&P Capital IQ, Moody's, PitchBook, and IBISWorld require existing subscriptions to those platforms. They provide the API connection; you still need the data license.
3. Microsoft 365 integration — Claude add-ins for Excel, PowerPoint, and Word are generally available; Outlook integration was announced as coming soon as of May 2026.

What the Templates Don't Cover

The 10 templates are deliberately broad — they cover the highest-frequency workflows at institutional buy-side and sell-side firms. Several important categories are outside their scope:

Specialized Regulatory Frameworks

Basel IV capital calculations, FRTB, IRRBB, CECL provisioning, IFRS 17 insurance, SA-CCR counterparty credit risk, NGFS climate scenario stress testing — these require domain-specific tools with verified calculation accuracy. A general "Statement Auditor" template is not designed to check whether a Tier 1 capital ratio calculation correctly applies the SA-CCR exposure methodology. That's where eval-tested specialized tools matter.

Private Credit and Alternative Markets

The templates are oriented toward public markets (earnings, public company data, sell-side research). Covenant monitoring for direct lending portfolios, LBO underwriting for PE-backed credits, DSCR analysis for infrastructure debt, and private fund administration are not addressed.

Tax, Accounting Standards Compliance

The GL Reconciler and Month-End Closer cover workflow mechanics, not technical accounting: ASC 842 lease accounting, ASC 326 CECL, ASC 740 tax provision, ASC 606 revenue recognition, FAS 133 hedging designations. These require step-by-step guided implementation against the specific standard, not a general reconciliation agent.

Free-Tier and API-Direct Users

Teams using Claude through the API directly, on free tiers, or building their own internal tools cannot access these templates. The independent tool ecosystem — open MCP servers, downloadable skills — is the primary resource for this segment.

How to Think About Eval Quality

Anthropic's announcement referenced Claude Opus 4.7's performance on the Vals AI Finance Agent benchmark — 64.37%, the top score among frontier models as of May 2026. That benchmark measures general finance reasoning accuracy across a standardized task set.

This is a different measurement than what ClaudeFinanceLab's eval framework tests. Our 66/66 eval score measures whether specific MCP tools return the mathematically correct answer for a defined calculation — Black-Scholes options pricing, WACC derivation, DCF valuation, portfolio VaR, credit migration matrices. These are closed-form tests with one correct answer, verified independently.

Neither benchmark alone is sufficient. Vals AI tells you how well Claude reasons about finance; ClaudeFinanceLab evals tell you whether a specific tool implementation is numerically accurate. Both matter — and the industry needs more public eval infrastructure, not less.

Recommended Approach: Templates + Independent Tools

For teams with Claude Enterprise and existing subscriptions to FactSet, S&P, or Moody's, the official templates are the fastest path to workflow automation for common institutional tasks. Start with the Earnings Reviewer (lowest configuration complexity) and the Month-End Closer (highest time savings).

For specialized technical calculations, regulatory reporting, private markets analysis, or non-Enterprise deployments, layer in independent tools from the finance MCP server directory and skills marketplace. The two ecosystems are complementary, not competing.

Finance MCP Servers Mentioned in the Announcement

All providers mentioned in the Anthropic finance agents announcement are now listed in the ClaudeFinanceLab directory with connection details, category tags, and auth requirements. Browse the full list at Market Data MCP servers, Data & Research MCP servers, and Risk MCP servers.

Related Resources

• Best MCP Servers for Finance (Full Directory)
• ClaudeFinanceLab Eval Results: 66/66 Passing
• Getting Started with Claude Finance Skills
• MCP Finance Server Setup Guide