Anthropic Financial Services Plugins Review in 2026
9 min read

Anthropic's financial services plugins are a free, Apache 2.0 licensed GitHub repo of ten Claude agents, vertical skill bundles and twelve data connectors for banking, research, private equity and fund admin. They run in Cowork or Claude Code on any paid plan, or headless via Managed Agents at token rates plus $0.08 per session-hour, and every output is a draft for human review.
What this covers
- The repo had 41,169 GitHub stars on 9 November 2026 and is licensed Apache 2.0, which permits commercial use and modification.
- It ships ten named agents, six vertical plugins plus LSEG and S&P Global partner plugins, and twelve MCP data connectors.
- Plugins run in Claude Cowork or Claude Code on paid plans; Managed Agents cost token rates plus $0.08 per session-hour.
- Connectors do not include data entitlements; FactSet, PitchBook and other providers need their own subscriptions.
- Agents draft work only: they do not post to ledgers, execute trades or approve onboarding.
- The cookbooks pin claude-opus-4-7 and subagent delegation is a research preview limited to one level.
Anthropic's financial services plugins are an open source set of Claude agents, skills and data connectors published at github.com/anthropics/financial-services under the Apache 2.0 license. They give Claude ready-made workflows for investment banking, equity research, private equity and fund administration, and every output is a draft staged for a human to review and sign off.
The repository had 41,169 GitHub stars and 5,734 forks when we checked it on 9 November 2026. It was created on 23 February 2026 and last pushed on 28 October 2026. This review covers what is in it, what it does, what the license allows, what it costs to run, and where it fits for a fund, family office or deal team.
What anthropics/financial-services is
The README describes the repo as "reference agents, skills, and data connectors" for the financial services workflows Anthropic sees most. Everything ships two ways from one source. You can install it as a plugin in Claude Cowork or Claude Code, or deploy it through the Claude Managed Agents API behind your own workflow engine, with the same system prompt and skills either way.
Anthropic announced the ten agent templates on 5 May 2026 and says they can be used as plugins in Cowork or Claude Code on all paid plans, or as Managed Agents on the Claude Platform, which it describes as in public beta. The repo itself is plain files. The README says everything is markdown and JSON with no build step, so you can read every instruction before you decide what data it is allowed near.

AI for finance teams
Screen more deals without adding analysts
First system live in about six weeks. You own the code.
Book a discovery callWhat is actually in the repo
There are three layers. Named agents own a workflow end to end. Vertical plugins carry the underlying skills and slash commands, grouped by business line. Connectors are MCP servers that wire Claude to data providers.
| Function | Agent | What the README says it does |
|---|---|---|
| Coverage and advisory | Pitch Agent | Comps, precedents and LBO into a branded pitch deck |
| Coverage and advisory | Meeting Prep Agent | Briefing pack before every client meeting |
| Research and modeling | Market Researcher | Sector or theme into an industry overview, landscape, peer comps and ideas shortlist |
| Research and modeling | Earnings Reviewer | Earnings call and filings into a model update and note draft |
| Research and modeling | Model Builder | DCF, LBO, 3-statement and comps, live in Excel |
| Fund admin and finance ops | Valuation Reviewer | Ingests GP packages, runs the valuation template, stages LP reporting |
| Fund admin and finance ops | GL Reconciler | Finds breaks, traces root cause, routes for sign-off |
| Fund admin and finance ops | Month-End Closer | Accruals, roll-forwards, variance commentary |
| Fund admin and finance ops | Statement Auditor | Audits LP statements before distribution |
| Operations and onboarding | KYC Screener | Parses onboarding documents, runs the rules engine, flags gaps |
The vertical plugins are financial-analysis (the core bundle with comps, DCF, LBO, 3-statement, deck QC and Excel audit), investment-banking, equity-research, private-equity, fund-admin and operations. Partner-built plugins from LSEG and S&P Global sit alongside them. The private equity bundle alone lists commands for deal sourcing, deal screening, diligence checklists, management meeting prep, unit economics, returns, IC memos, portfolio monitoring, value creation plans and AI readiness.
The connector file in the core plugin declares twelve MCP servers: Daloopa, Morningstar, S&P Global, FactSet, Moody's, MT Newswires, Aiera, LSEG, PitchBook, Chronograph, Egnyte and Box. The README notes that MCP access may require a subscription or API key from each provider.
What it actually does
In Claude Code, you add the marketplace with claude plugin marketplace add anthropics/financial-services, install the financial-analysis core first, then add the agents and verticals you want. After that, skills fire when Claude judges them relevant and slash commands such as /comps, /dcf, /earnings and /ic-memo are available in the session. In Cowork you paste the repo URL into the plugin settings or upload a zipped plugin folder.
The Managed Agents route is where the engineering is most visible. Each of the ten cookbooks has an agent.yaml, a set of leaf-worker subagents and steering-event examples. A deploy script uploads the skills, creates the workers and posts the orchestrator to the agents endpoint. In every cookbook only one worker holds the Write tool.
The GL Reconciler shows the security pattern clearly. Its reader worker touches untrusted counterparty and custodian statements but has only Read and Grep and no connectors. The orchestrator holds read-only GL and subledger connectors but never opens outsider files. A critic re-verifies each break against trusted sources, and a resolver writes the exception report without ever opening an external document. The cookbook states plainly that none of this writes to a system of record and that ledger adjustments need human approval outside the agent. For any firm designing its own agents, this separation is worth copying even if you never install the plugins.
The license and what it allows commercially
The repository is licensed under Apache 2.0. The license text grants a perpetual, worldwide, royalty-free copyright license to reproduce, prepare derivative works of, sublicense and distribute the work, plus a patent license. In practice a firm can fork the skills, rewrite them to its own conventions, and run them inside a commercial operation without paying Anthropic for the files.
Three boundaries matter. The license grants no rights to Anthropic's trademarks beyond describing where the work came from. The work is provided as is, with no warranty, and you carry the risk of using it. And the license covers the repo only: it does not give you Claude usage, and it does not give you data from FactSet, PitchBook or any other provider whose connector appears in the config.
The README adds its own disclaimer. Nothing in the repo is investment, legal, tax or accounting advice, and the agents do not make investment recommendations, execute transactions, bind risk, post to a ledger or approve onboarding.

Build library
See what we have already shipped
Working systems, the stack behind them, and what they cost to run.
Open the libraryWhat it costs to run
The files are free. Running them costs three things: Claude access, model usage if you go headless, and data.
For the plugin route, you need a paid Claude plan. On 9 November 2026, claude.com listed Pro at $20 per month billed monthly ($17 per month on annual billing), Team standard seats at $25 per seat per month billed monthly ($20 on annual), Team premium seats at $125 billed monthly ($100 on annual), and Enterprise at $20 per seat per month plus usage at API rates, billed annually with a 20 seat minimum. Plan usage limits apply, and heavy agent runs draw from the same pool as chat.
For the Managed Agents route, you pay standard token rates plus $0.08 per session-hour of active runtime. The cookbooks we checked (GL Reconciler, Pitch Agent and Model Builder) pin claude-opus-4-7, which the pricing page lists at $5 per million input tokens and $25 per million output tokens. Opus 5.5 is listed at $4 and $20, so updating the model line is a cost decision as well as a quality one. Code execution is free for 50 hours a day per organization and $0.05 per container hour after that.
Data is the line item most teams underestimate. The connectors are wiring, not entitlements. If you do not already license a provider, its connector does nothing for you, and the repo does not publish what any of those subscriptions cost. Most firms will also need to stand up their own read-only MCP endpoints for internal systems; the GL Reconciler cookbook expects you to supply GL_MCP_URL and SUBLEDGER_MCP_URL yourself.
Limits to know before you rely on it
These are reference templates, and the README says so: "They're starting points." There is no settings screen and no service level. The value comes from someone on your team tuning the prompts, skills and connectors to how your firm works.
Subagent delegation through callable_agents is a research preview and supports one delegation level. The deploy script also sends a Managed Agents beta header, so plan for an API surface that may still change.
Output quality needs checking. Anthropic's launch post cites Claude Opus 4.7 at 64.37% on Vals AI's Finance Agent benchmark, the leading score at the time. That is strong for the field and still a clear reason every model, memo and reconciliation goes past a qualified reviewer.
The repo also shows some housekeeping drift, which is normal for a fast-moving project but worth knowing. The README says the core plugin carries "all 11 data connectors" while the config declares twelve. Grand Linux Solution reported on 5 September 2026 that the core .mcp.json was not valid JSON, so the connectors did not load after install. A commit on 28 October 2026 titled "Fix malformed JSON in financial-analysis .mcp.json" addressed it, and the file parsed cleanly when we fetched it. The financial advisors plugin was removed from this repo in September 2026 and now lives at its own repository, anthropics/claude-for-financial-advisors, but the README table still links the old folder, which returns not found.
Finally, the fund-admin skills are written around fund practice. Grand Linux's review notes that skills such as accrual schedules and NAV tie-outs use LP and GP vocabulary and carry no local tax or statutory chart of accounts logic, so operating companies outside that frame will need to add their own context.
Who should use it and who should not
A good fit
Teams already on a paid Claude plan with someone comfortable editing markdown and YAML. Analysts who want a fast start on comps, DCFs, CIM drafts or IC memo skeletons and will review every output. Platform teams that want a well-built reference for agent permissions before designing their own. Firms that already license some of the connected data providers.
Not yet a fit
Teams that want software that works out of the box with no editing. Workflows where the agent would need to post entries, approve onboarding or act without sign-off, which the repo explicitly rules out. Firms whose key data sits in internal systems with no MCP endpoint and no one to build one. Anyone who cannot accept a beta API surface for headless runs.

Implementation services
Put AI into the workflow you actually run
Diligence, close, reconciliation, reporting — in your environment.
See how it worksWhere it fits for a fund, family office or deal team
For a private equity deal team, the private-equity and investment-banking bundles cover the document-heavy front end: screening an inbound CIM, building a diligence checklist, drafting the IC memo. That is useful scaffolding, and it maps closely to the workflows we cover in AI for private equity firms and AI for financial due diligence. The gap is your own process: your screening criteria, your memo template, your data room and your CRM.
For a family office, the fund-admin agents are closest to the daily pain. Statement Auditor and Valuation Reviewer target LP statements and GP packages, which is exactly the paperwork a multi-manager office receives. The question is whether your statements, custodians and accounting system can be reached through connectors, which is the integration work described in our page on AI for family offices.
For an advisory or deal desk, Pitch Agent and Meeting Prep Agent show the shape of a working system. Turning that into something that runs every day against live deals is the subject of our AI deal desk systems work.
When an in-environment build is the better answer
The plugins are a strong starting point when the workflow is generic and the data is already licensed. They are a weaker fit when the value sits in your firm's own judgment, templates and internal systems, when the agent must run inside your cloud under your controls, or when nobody in-house has time to maintain a fork as the upstream repo changes.
That is the alternative to buying another seat. Consultance.ai builds custom finance AI inside the client's environment, with the first system live in about six weeks, and the client owns the code. The Anthropic repo is often where we start, because the security pattern is sound and the license allows it. If you want to see working templates first, browse the consultance.ai prompt library, or book a call to talk through a specific workflow.
Sources
- anthropics/financial-services README and repository metadata, GitHub (fetched 9 November 2026)
- Managed agent cookbooks README and GL Reconciler cookbook, GitHub (fetched 9 November 2026)
- financial-analysis .mcp.json connector config, GitHub (fetched 9 November 2026)
- Apache License 2.0 text in the repository, GitHub (fetched 9 November 2026)
- Repository commit history, GitHub (fetched 9 November 2026)
- anthropics/claude-for-financial-advisors repository, GitHub (fetched 9 November 2026)
- Agents for financial services, Anthropic (fetched 9 November 2026)
- Plans and pricing, Claude (fetched 9 November 2026)
- What Is Claude for Financial Services?, Grand Linux Solution (fetched 9 November 2026)
