Financial Services: Claude agents and skills for financial work

Everything worth knowing about anthropics/financial-services: a collection of plugins, skills, reference agents, and MCP connectors for financial analysis and operations tasks with Claude.

---

What Financial Services is

Financial Services is an Anthropic repository with reference material for configuring Claude Code for investment banking, equity research, private equity, wealth management, fund administration, and operations. It isn’t an autonomous system that executes transactions: it offers plugins for Claude Cowork and Claude Code, plus examples of API-managed agents.

The declared scope includes DCF models, comparables, LBOs, three-statement models, earnings review, meeting prep, general ledger reconciliation, KYC controls, valuations, and month-end close. The repository explicitly warns that the output is working material for review by qualified professionals; it doesn’t constitute investment, legal, tax, or accounting advice, and agents shouldn’t execute trades, approve onboarding, or post journal entries.

The origin: an Anthropic vertical library

GitHub records the repository’s creation on February 23, 2026. The owning organization is Anthropic; the repository presents itself as a reference implementation of its agents, skills, and connectors for financial services.

The offering answers a practical tension: general-purpose assistants can draft or summarize, but a financial workflow requires templates, data sources, control rules, and explicit delimitation of authority. That’s why it combines domain instructions with MCP connectors and, in the managed examples, separates document reading, orchestration, and resolution actions.

No official launch announcement was recovered that would establish a public announcement date distinct from the GitHub creation date. The first direct thread found on Hacker News appeared on May 7, 2026, but was submitted by srikanth235; it establishes reach for the repository, not an independent review.

Philosophy and principles

• Operational specialization: plugins are organized by vertical and by complete workflows, rather than offering a single generic financial assistant.
• Accountable human: responses are framed as preparatory work requiring professional review, not authorization to move money, accept risk, or modify a ledger.

• Scope-limited data and tools: the gl-reconciler example uses read-only MCP for orchestration, restricts untrusted documents to reading and search, and reserves writing for a resolver component.

• Reuse of instructions: the same skills and prompts are packaged as Cowork plugins and as managed agent examples.

How it works

The repository’s marketplace declares 20 plugins: seven verticals, eleven workflow agents, two data-partner plugins, and one Microsoft 365 installer tool. Verticals include financial-analysis, investment-banking, equity-research, private-equity, wealth-management, fund-admin, and operations; agents include pitch-agent, earnings-reviewer, model-builder, gl-reconciler, kyc-screener, and month-end-closer.

The financial-analysis plugin is designed to connect with providers like Daloopa, Morningstar, S&P Global, FactSet, Moody’s, MT Newswires, Aiera, LSEG, PitchBook, Chronograph, Egnyte, and Box; required subscriptions or keys depend on each provider. The partners included in the manifest are LSEG and S&P Global.

For the managed agent examples, each recipe incorporates agent.yaml, depth-one subagents, and steering-examples.json. The deployment script resolves local prompts and skills, loads the skills, creates subagents, and requests the agent from the API.

Official and semi-official status

The repository is official Anthropic, and its README documents installation as a Claude Code plugin marketplace and as a Claude Cowork plugin. In practice, that means the instructions are distributed from the vendor’s own repository, not that every data connection, financial output, or control policy is certified for a specific regulated use case.

Its status is also semi-official regarding partners: LSEG and S&P Global appear as plugins included in the marketplace manifest. That listing establishes that the repository packages them; it isn’t sufficient on its own to infer regulatory certification or a recommendation of all their configurations.

The ecosystem

Related Anthropic repositories

An organization search places this project alongside other skill and plugin repositories: anthropics/skills (169,263 stars) and anthropics/claude-plugins-official (33,516).

Also present are anthropics/knowledge-work-plugins (23,473) and anthropics/claude-for-legal (9,095). These are sibling repositories from the same publisher; figures were checked on August 14, 2026 and don’t prove they’re technical dependencies of Financial Services.

Forks, ports, and community extensions

• NBreeze-Eric/financial-services-plugins-cn is the highest-starred fork returned by the queried API: 8. Its description declares an adaptation for Chinese stock research and iFinD.
• jwangkun/claude-for-financial-services-cn declares adapting the official project for working with Chinese stocks and had 710 stars when checked. It’s a community adaptation in Chinese, not an Anthropic repository.
• JoelLewis/finance_skills declares 81 financial services skills distributed across seven plugins and had 169 stars. It’s an adjacent extension or alternative, not an official fork.

These derivatives shouldn’t be confused with Anthropic support: the forks API proves the fork relationship, while the external repositories’ descriptions describe their own goals.

Repo numbers

Measured: August 14, 2026, GitHub API.

Metric	Value
Stars	34,265
Forks	5,105
Real subscribers	266
Commits in main history	64
Open issues per API	189
Primary language	Python
License	Apache-2.0
Created	February 23, 2026
Last push to main	August 4, 2026
Releases and tags	No releases or tags recovered

The top contributors returned by the API were mihilmy (18 contributions), cxyback (17), manar-ant (12), and tobinsouth (5).

GitHub reports watchers_count the same as stars in the general response; that’s why the table uses subscribers_count as the measure of real subscribers.

Likewise, open_issues_count may mix issues and open pull requests; it shouldn’t be interpreted as an issues-only count.

The releases API returned no versions and the repository has no tags recovered in this measurement.

How to contribute

No CONTRIBUTING.md, pull request template, issue template, or code of conduct was recovered: GitHub’s community profile marks 25% completeness. The available CLAUDE.md guide does describe a technical workflow for skill changes, though: edit plugins/vertical-plugins/<vertical>/skills/, sync the packaged copies, and run the following checks.

python3 scripts/sync-agent-skills.py
python3 scripts/check.py

check.py validates manifests and cross-references, detects mismatches between packaged skills, and can install core.hooksPath=.githooks inside a Git clone. Because of that side effect, it’s not worth running solely for inspection. The same guide requires avoiding non-ASCII content in BOM-less PowerShell files, since Windows PowerShell 5.1 can misinterpret those files.

Quick-start guide

Installation and first run

As a prerequisite, Claude Code is needed for the terminal route. The README says to add the marketplace and install only the needed plugins; for example:

claude plugin marketplace add anthropics/financial-services
claude plugin install financial-analysis@claude-for-financial-services
claude plugin install pitch-agent@claude-for-financial-services
claude plugin install gl-reconciler@claude-for-financial-services

In Claude Cowork, the alternative first run is to open Settings → Plugins → Add plugin, paste the repository URL, or upload a ZIP of a folder under plugins/. To deploy a recipe as a managed agent, ANTHROPIC_API_KEY is needed, and the README shows this pattern:

export ANTHROPIC_API_KEY=sk-ant-...
scripts/deploy-managed-agent.sh gl-reconciler

Common workflows

1. Building a financial analysis: install financial-analysis and request the model, comparables, LBO, or presentation review the plugin covers.
2. Preparing a deal pitch: install pitch-agent; its declared flow goes from comparables, precedents, and LBO to a branded presentation.
3. Reconciling general ledger and subledger: install or deploy gl-reconciler; it requires GL_MCP_URL and SUBLEDGER_MCP_URL to access its MCP connectors.
4. Updating earnings work: install earnings-reviewer, whose declared flow goes from calls and regulatory filings to model updates and a draft note.

Essential configuration

• .claude-plugin/marketplace.json: the catalog that decides which plugins, versions, and paths get published.
• plugins/vertical-plugins/<vertical>/skills/: the location of skills a contribution should edit and then sync.
• plugins/vertical-plugins/financial-analysis/.mcp.json: the manifest of financial analysis’s data connectors.
• GL_MCP_URL and SUBLEDGER_MCP_URL: variables required by the managed gl-reconciler recipe.
• managed-agent-cookbooks/<agent>/agent.yaml: the definition of a managed agent recipe.

Common pitfalls and fixes

• Invalid MCP manifest: validating the current financial-analysis file with jq empty fails on line 47 because a comma is missing between the egnyte and box configurations. Until it’s fixed upstream, review and repair that separator before using a JSON validation or loading that manifest.

• A provider without credentials: a connector being listed in the plugin doesn’t remove the provider’s subscription or API requirements. Configure the credentials and permissions each provider requires before expecting results.
• Confusing preparatory work with an authorized decision: keep professional review and don’t use the agent for trades, risk approvals, journal entries, or onboarding, per the README’s limitations.
• Running checks with Git side effects: scripts/check.py can change the local core.hooksPath value; run it in a working copy prepared for that.

Integrations and migration

The project integrates MCP connectors for financial data and an installer for Claude access from Microsoft 365. To migrate a custom skill into the repository’s packaging, the documented path is placing it in the corresponding vertical directory, syncing the agent copies, and passing the checks. No guide was found describing automatic migration from another agent platform.

How the community received it

The recoverable external reception is limited and should be read with caution:

• Hacker News thread 48049207, submitted by srikanth235 and linked directly to the repository, had 3 points and 0 comments. It’s a reach signal, not a review or a sign of satisfaction, because there were no comments.
• In another Hacker News thread, yepyoukno explicitly recommended Anthropic’s financial skills, called them a curated knowledge distillation, and linked the repository. It’s a favorable individual opinion, not a comparative evaluation or a consensus.
• The 66-second video “Unlock Financial Insights with Anthropics’ GitHub Repo Review!”, from Simple Tech Lab, showed 50 views and describes the structure and utilities as promising. Its length and scale limit its value as technical analysis.
• The video “GitHub - anthropics/financial-services”, from GitHub Daily Trend AI Podcast, showed 37 views and no visible rating; its format is trend aggregation, not an independent review.

No recoverable evidence was obtained from Reddit due to an access page, from X due to a login redirect, or from Product Hunt due to a verification challenge. An exact npm package page also wasn’t recovered; these access limits don’t prove an absence of conversations or packages.

Financial Services versus other approaches

Approach	Verifiable overlap	Verifiable difference
jwangkun/claude-for-financial-services-cn	Adapts Anthropic’s project to financial tasks.	It declares an orientation toward Chinese stocks; Financial Services is Anthropic’s original repository.
JoelLewis/finance_skills	Offers financial services skills in plugins.	Its README declares 81 skills across seven plugins; it isn’t presented as an official fork.
anthropics/claude-for-legal	An Anthropic sibling repository for a regulated professional domain.	It’s oriented toward the legal field, while Financial Services publishes financial analysis and operations workflows.

The right choice depends on jurisdiction, data sources, internal policies, and human review. This repository provides templates and reference connectors; it doesn’t replace model controls, permissions, or an organization’s own compliance processes.

Use cases and who this repository can help

• Investment banking and equity research teams can use financial-analysis, pitch-agent, earnings-reviewer, and model-builder as a starting point for modeling, comparables, presentation, and earnings tasks.
• Fund operations and accounting can study gl-reconciler, month-end-closer, and the separation controls between reading, orchestration, and resolution documented in the reconciliation recipe.
• Compliance and onboarding operations teams can start from kyc-screener and the operations vertical, always keeping the decision limits and human review the project indicates.
• Internal agent developers can reuse the plugin, skill, MCP manifest, and managed agent recipe structure to connect data sources already authorized by their organization.

Resources

• Repository: https://github.com/anthropics/financial-services
• Documentation and installation: https://github.com/anthropics/financial-services/blob/main/README.md
• Official plugin manifest: https://github.com/anthropics/financial-services/blob/main/.claude-plugin/marketplace.json
• Technical contribution guide: https://github.com/anthropics/financial-services/blob/main/CLAUDE.md
• Managed agent recipe: https://github.com/anthropics/financial-services/tree/main/managed-agent-cookbooks/gl-reconciler
• Discussions: https://news.ycombinator.com/item?id=48049207, https://news.ycombinator.com/item?id=48342105
• Videos: https://www.youtube.com/watch?v=L30V4D1SAqA, https://www.youtube.com/watch?v=6vDGQUthHSY
• Chinese adaptation: https://github.com/jwangkun/claude-for-financial-services-cn

---

Note: this article combines the financial-services README, the GitHub API, the CLAUDE.md guide, Hacker News results, and YouTube videos retrieved on August 14, 2026. Figures change over time.