Claude for Financial Services
github.com/anthropics/financial-services- Category
- AI Agents
- Rank
- No. 29Tools index
- Type
- APP
- Builder
- anthropics
- GitHub
- 34.8k stars
- Date
About
Anthropic's open plugin suite bringing Claude into financial-services workflows.
What it does
A set of ready-made financial workflow agents, domain skills, explicit commands, and data connectors. Teams can run complete jobs such as pitch preparation, earnings review, model building, reconciliation, and KYC screening, or install narrower capabilities for specific tasks. The same prompts and skills support interactive use and deployment behind a firm's workflow engine.
Why it's ranked here
The strongest case is practical breadth joined to a coherent delivery model. Self-contained agents cover research, modeling, operations, and onboarding, while narrower bundles let teams adopt only the needed methods and connectors. Shared source material reduces divergence between interactive and managed deployments. However, subagent delegation remains a research preview, and every work product still requires qualified human review.
What's good
The templates produce recognizable analyst work products rather than generic chat responses. Firms can tune terminology, procedures, branded presentation layouts, agent scope, and data sources. Skills are authored centrally and copied into self-contained agents, with checks that detect drift. Managed deployments include reference orchestration, payload validation, target allowlisting, security notes, and dry-run checks. Microsoft 365 tooling also supports tenant provisioning and per-user access rules.
Tradeoffs
This is reference infrastructure, not an autonomous financial decision system. It neither recommends investments nor executes transactions, posts ledger entries, approves onboarding, or removes compliance responsibility. Managed orchestration is explicitly a reference implementation, and quoted document content can imitate handoff messages despite the included mitigations. Structured output needs harness-side validation. Many data connectors may require separate subscriptions or API keys, while production workflows still need firm-specific prompts, controls, and systems integration.
How to use it well
It suits financial institutions that already have qualified reviewers and want repeatable first drafts for models, memos, research notes, reconciliations, reporting, or screening. Start with one bounded workflow, connect approved data sources, add firm conventions and templates, then keep human sign-off explicit. Use managed deployment when an existing workflow engine should control execution. Do not treat it as investment advice, a transaction platform, a ledger, or a substitute for compliance governance.
Technical notes+
The repository is primarily file-based Markdown, JSON, and YAML with no build step. plugins/agent-plugins/ contains self-contained workflow plugins, while plugins/vertical-plugins/ holds canonical skills, commands, and MCP configuration. managed-agent-cookbooks/ supplies agent.yaml, depth-1 workers, steering examples, and security guidance. scripts/check.py parses manifests, resolves references, compares bundled skills with their sources, checks required cookbook files, and gates PowerShell encoding. scripts/sync-agent-skills.py replaces bundled copies from vertical sources. scripts/validate.py applies JSON Schema outside the API. scripts/orchestrate.py parses handoff text, allowlists targets, validates payloads, and warns that typed events would be safer. scripts/test-cookbooks.sh dry-runs deployments. claude-for-msft-365-install/scripts/build-manifest.mjs customizes Office and Outlook manifests, while claude-for-msft-365-install/examples/python-bootstrap/app.py demonstrates Entra token validation and rule-based distribution of skills and MCP servers.
Observed
- License
- Apache License 2.0
- Content and tooling
- File-based Markdown, JSON, and YAML with Python, shell, and JavaScript administration scripts; no build step
- Install surface
- Claude Cowork repository or ZIP installation, Claude Code marketplace installation, and Managed Agents API deployment
- Interfaces
- Self-contained agents, automatic skills, slash commands, MCP connectors, and Managed Agent templates
- Data integration
- MCP connectors are centralized in the financial-analysis core plugin; provider access may require subscriptions or API keys
- Microsoft platform support
- Separate administration tooling provisions the Claude add-in for Excel, PowerPoint, Word, and Outlook
- Deployment structure
- Managed-agent cookbooks use an orchestrator with depth-1 leaf-worker subagents and external schema validation
Read from README.md, scripts/check.py, scripts/validate.py, scripts/orchestrate.py, scripts/version_bump.py, scripts/sync-agent-skills.py, claude-for-msft-365-install/scripts/build-manifest.mjs, claude-for-msft-365-install/examples/python-bootstrap/app.py, claude-for-msft-365-install/examples/python-bootstrap/config.py, claude-for-msft-365-install/examples/python-bootstrap/get_tenant_id.py, claude-for-msft-365-install/examples/python-bootstrap/mint_dev_token.py, LICENSE, CLAUDE.md, .githooks/pre-commit, scripts/test-cookbooks.sh.
Intel on Claude for Financial Services
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.