I read the Claude for Financial Advisors repo to see what it really does, and what it doesn't.

"A set of ready-to-run workflows for Claude Cowork…"

"Reference implementation. Not actively maintained or monitored, and not accepting contributions."

Both lines come from the same README, for the plugin Anthropic launched for financial advisors this week. So which is it?

*You can find the Anthropic Claude for Financial Advisors Github repo **here*

I got some version of that question all week, at Future Proof and in my inbox. A few firms were convinced it changed what they were building or planning right now. I held off on a real opinion until I was home and could read the actual skills, agents and connector tools. In almost every case I've looked at, the impact on what those firms are doing today is close to zero.

And yes, I'm writing about it because everyone is paying attention. At least I'm telling you that up front.

TLDR: This is a directional announcement about where Anthropic and these platforms are going, dressed up as a ready-to-implement solution for any advisor.

What actually shipped

  • Eight skills: onboarding, pre-meeting prep, post-meeting follow-up, a compliance pre-check, prospect intake, rebalance review, an alternatives brief, and an estate and tax brief.
  • Six small helper agents that do narrow jobs for those skills, like pulling data from one system or checking a statement.
  • A list of connector addresses pointing to partner systems.
  • All of it is plain text files. The README describes the skills as "written as plain instructions." There's no software code doing the work.
  • No new AI model, no portfolio engine, and no client data stored. The connectors are built and run by the partner firms. Claude calls them and works with what comes back.

That's real work, and the partner access is the valuable part. But what you're getting is a framework. If you add the connectors and run the skills as shipped, you'll get the same output as the next firm that does the same thing.

The biggest name isn't connected yet

Charles Schwab's press release calls Schwab "currently the only RIA custodian providing integration with Claude" through the new offering. The repo lists Schwab as "coming soon" and says plainly: "The connector is not yet available."

That matters more than it sounds. Of the roughly 11,200 SEC-registered advisers that list a custodian on Form ADV, about 7,100, nearly two in three, list just one. (Form ADV only asks firms to list custodians holding 10% or more of their separately managed account assets, so "one" means one primary custodian.) That's according to RIA Ascent (riaascent.com), based on SEC Form ADV filings as of August 2026.

Article content

Schwab connector is not available in Claude despite the press release (as of September 18, 2026)

If your primary custodian isn't Schwab, there's no custody connector in this launch. If it is Schwab, there isn't one yet either.

Article content

Charles Schwab Custodian Details courtesy of RIAAscent.com

I expect the Schwab connector to arrive soon. But it's a significant gap, given how much of this week's excitement was built on it.

A logo isn't a data feed

I copied the Orion tool list from the Orion connector page in the Claude desktop app on September 18.

The Orion Connect tools find accounts and households, list a household's accounts, look up owners and addresses, return current value, and run reports. None of them hands back holdings, performance or drift directly. Anthropic's own pre-meeting skill notes that running a report returns a link that expires after 24 hours, and it tells Claude not to pull numbers out of that link. The Redtail side can create notes, activities and opportunities.

Article content

Orion MCP details from Orion.com

That's a reasonable first release, and it may be a deliberate choice on Orion's part. But "Orion connector" on a launch slide and portfolio numbers in a meeting brief are not the same thing.

Addepar's connector, by contrast, returns holdings, performance, transactions and private fund activity directly. The difference is a choice each vendor made, not a limit of Claude.

The marketing also gets ahead of the product. Anthropic's blog says the post-meeting skill produces "a client summary, a recap email, and CRM tasks in one pass." The skill itself says it is "not drafting an email for the advisor to send," and it lists sending anything to the client as out of scope. Both describe a reasonable direction. They just describe two different products.

The guardrails are instructions

The README is candid about this. A few safeguards are built into the software. The rest, including getting the advisor's approval before anything is written back, are "enforced by the model following them, not by the runtime." There are real controls in the admin settings, and setting them up is the firm's job.

The risk I'd spend the most time on is pulling the wrong client's data.

Most firms create extra records in their portfolio, reporting and trading systems to work around what those systems can't do on their own. There's a family household that includes the kids, so reports show total family wealth. Then the kids and the trusts get their own separate records so they can be traded, rebalanced and billed. None of that is bad data that requires cleanup. It's intentional and needed to get the work done.

In the application, the screen made sure you picked the right record before you started. Now that screen is gone. The prompt has to find the right record on its own, and when a family has four legitimate records, it's easy to grab the wrong one.

The skills do ask the advisor to confirm the household. But they also tell Claude to suggest the "clearly better match," such as the most recently active record. For a family review, the most recently active record can easily be the one set up for trading, not the one set up for reporting. Anthropic's own skill calls pulling the wrong household's data "a privacy incident."

I AGREE!

Who can use it today

Claude Enterprise starts at $4,800 a year, which covers 20 seats. That's before any usage. Anthropic's Peter Nolan put the ballpark for an active user at $70 to $120 a month. To be fair, it's not clear whether he meant that to include the Enterprise subscription or sit on top of it. Assuming it sits on top, 20 users would add roughly $16,800 to $28,800 a year in usage, for a launch that is mostly read-only.

Article content

Claude Teams and Enterprise pricing as of September 18, 2026 via Claude.com

Treat that as a starting cost. As the ability to write back to systems arrives, usage and cost will rise with it. The added value may well justify it. It's a factor to budget for, not a criticism. It also shows why this makes sense for Anthropic.

The harder problem is access. The firms that could benefit most are the smaller, single-custodian RIAs. Their data sits in fewer places, and they want to use AI safely but don't know where to start. They're also the least likely to have 20 seats. And the cheaper plans give up the compliance tooling, because audit logs come only with Enterprise. The firms that could use this to compete with firms ten times their size are the ones it's hardest for to reach.

Why it still matters

I want to be fair here. The skills are thoughtfully built. They label every number as either read from a system or worked out by Claude. They do the math with a calculator tool instead of in the model's head. And when two systems disagree, they stop and ask rather than guess. That's better than most of the prompt libraries I've reviewed.

Note: I wanted to comment on the "calculator tool" sentence above...this is a bid deal! It's a reminder that these systems are probabilistic and you still need deterministic code for mathematical calculations. This is something a lot of advisors turned coders need to remember when trying to replace their tools with AI. I digress...please read on.

The bigger shift is outside Claude itself. Vendors are being pushed to open up their data through controlled connections instead of keeping it locked inside their own apps. William Trout of Datos Insights suggested that if more custodians connect, this could become the standard way advisor systems plug into each other. A launch partner's CEO called it "the most important thing to happen to our industry in a decade," then pointed out that exact tax and estate calculations still need calculation engines that give the same answer every time, not a language model.

Not everyone signed on. Jason Wenk told RIABiz.com that

"Anthropic had asked us to be the exclusive custody partner but it didn't make sense for us at this time," while also calling Claude a wonderful product.

I spoke with the CTO of another major platform that also chose to pass.

The vendors who did participate are in a curious position. By opening their workflows to Claude, they're showing exactly where those workflows are thin. Nolan said, "We aren't trying to replace any of the tools out there." I take him at his word. But I'll note that a statement on stage is still not a constraint on a company. That's my read, not a prediction.

What I'd tell a client

Don't just add the connectors and run the plugin as shipped.

Use the skills as a starting point, and review every one of them against your own households, account structures and procedures. Decide where it should run: on your own infrastructure, on Anthropic-managed agents, or on Claude Enterprise. Put the guardrails in the connector and admin settings, not just in the instructions. Remove any bloat with references and instructions to products you don't use and bring in purpose-built providers wherever they do the job better.

The eight skills mostly mirror how advisors work today. That's fine, as long as it's the starting point and not the destination.

Treat this plugin as a well-written first draft of your firm's AI operating procedures. The firms that get the most out of it will be the ones that finish writing it.

The full analysis, covering connector-by-connector status, the compliance and recordkeeping picture, and what a fair pilot looks like, is here: [link]. Link will be added shortly. You can view the full connector list directly from the Github readme file at the bottom of this page.

*If you want to run this for real, I'd like to partner with a firm to implement it and publish the results as a case study. We're also putting together a package to help firms implement and customize it. Reach out to *James Cantwell 🎯 *at james@wealthtechselect.com.*

Additional Info: Here's how Anthropic's README describes each connector:

Each skill pulls from whichever connectors are live and falls back to paste/upload when a system isn't connected. Every write to a client system pauses for advisor approval.

  • Addepar: Brings investment intelligence into Claude, giving advisors governed access to portfolio data, analytics and workflows across public and private markets.
  • BlackRock: Brings Advisor Center's portfolio analytics, model portfolios, and investment research resources to advisors through Claude.
  • SS&C Black Diamond Wealth Solutions: Brings portfolio, performance, holdings and rebalancing data into Claude, so advisors can review their book, spot clients out of tolerance, surface tax loss harvesting opportunities, and generate client reports, tasks and rebalance sessions with their approval.
  • Envestnet | Tamarac: Brings portfolio and performance reporting, including reconciliation status, from Tamarac into Claude so that advisors can conduct household reviews, drift checks, and meeting prep.
  • iCapital: Brings a client's alternatives book into Claude, with holdings and performance across private funds, so advisors can see NAV against commitments, unfunded capital, recent calls, and distributions alongside the liquid portfolio.
  • Envestnet | MoneyGuide: Brings a household's financial plan into Claude — net worth, probability of success, and goals.
  • Orion Advisor Solutions: Brings portfolio and performance reporting from Orion Connect, and client records from Redtail CRM, into Claude so that advisors can conduct household reviews, drift checks, and meeting prep.
  • Vanguard: Brings information on its model portfolios and advisor investment solutions into Claude, so advisors can draw on Vanguard's trusted research, portfolio construction, and asset allocation expertise when making decisions for clients.
  • Wealthbox : Connects Claude to client records and meeting history to power onboarding, meeting prep, and follow-up.
  • Wealth.com: Gives Claude a structured view of each client's estate plan, including trust and will summaries, and the full balance sheet, so advisors can begin document review using an organized summary of available information instead of a binder full of documents.
  • Zocks | AI for Advisors: Brings Claude the client intelligence it captures from every conversation, including profiles, goals, life events, and commitments.
  • Charles Schwab (coming soon): Custodial account, titling, and beneficiary data for the estate and rebalance reviews. The connector is not yet available; skills that reference Schwab note that and fall back to paste or upload until it lands.

These join other connectors already available in Claude, including Microsoft 365, Salesforce, Box, FactSet, S&P Global, Morningstar, and more.

Sources

Related: AI Isn’t Replacing Your CRM. It’s Replacing What Your CRM Was Built To Do.