How to Set Up Claude for Financial Advisors: Step by Step for RIAs

1. The Setup at a Glance

Installing Claude for Financial Advisors takes a few clicks. Setting it up so your CCO can sign off takes longer, because the slow parts are decisions and vendor access, and neither comes in the box.

The order below follows one rule: no real client record goes through a skill until the records and approval pieces are in place. If you are still deciding whether to use the plugin at all, start with the complete guide to Claude for Financial Advisors.

A solo advisor can skip most of step two. The records question in step four applies to everyone.

2. Before You Start

Get four things in hand before anyone clicks Add.

• The plan. Know whether you are buying Enterprise or Team, and why. Section 3 covers the choice.
• The Primary Owner. Find out who holds that role in your Claude organization. Owners manage organization plugins, and only the Primary Owner can switch on the Compliance API, which section 9 depends on.
• The desktop app. Cowork is the agentic mode in the Claude desktop app, so every pilot advisor needs the app installed and signed in to the firm's organization.
• A list of systems, with who holds each admin login. Write down every system a skill could read: CRM, portfolio platform, custodian, planning and estate software, meeting notes, email and calendar, and your archive. Next to each, note who administers it, whether your contract covers connector or API access, and whether the vendor offers a read-only connection.

That last list takes the longest, because the answers sit with vendors. Anthropic's install tutorial warns: "Access to these connectors may require a separate subscription or API key from the respective provider."

Better to learn that from a spreadsheet in week one than from a stalled pilot.

3. Buy or Confirm the License

Anthropic's launch post is direct: "We recommend Enterprise plans for registered investment advisers, because it includes the audit logs that support recordkeeping." Enterprise is also the only plan with the Compliance API, which your archive vendor needs in section 9.

Enterprise pricing, as InvestmentNews reported it from Anthropic's website, is a seat fee of $20 per user per month with a 20-seat minimum, plus usage billed separately at API rates. Anthropic's billing page adds that the seat fee includes no usage, and that usage is pooled across the organization.

Team is the cheaper route. Standard seats run $25 per member per month billed monthly or $20 billed annually, for 2 to 150 members, with Cowork, connectors, SSO and spend controls included, per the Team plan page. Team has no audit log export and no Compliance API.

Decide with your CCO and write the decision down with the reason. If you choose Team, your records procedure carries more weight later. The pricing guide has the numbers by firm size, and the small RIA guide covers the Team trade-off.

4. Turn On Cowork and Choose How the Plugin Reaches Advisors

Confirm that Cowork is available to everyone who will use the plugin. Team and Enterprise seats both include it.

Then pick how the plugin reaches each advisor:

• Personal install. Each advisor adds the plugin on their own. That suits a solo advisor or a first look. At a firm, it means each advisor may run a different copy, and nobody is sure which.
• The organization marketplace. On Team and Enterprise, Owners manage plugins in Organization settings, Plugins and skills, and give each plugin an install preference: Installed by default, Required (members cannot remove it), Available to install, or Not available. Members cannot edit organization-managed plugins. Enterprise can also set access by group, and when someone sits in two groups, the most permissive setting wins.

A firm should use the organization route. On Enterprise, give the pilot group access first, then set the plugin to Required for everyone at launch.

While you are in those settings, decide your policy on local MCP servers. On Enterprise, admins can restrict which plugins members install and can switch local MCP servers off, and Anthropic's plugin help page warns: "Plugins may include local MCP servers that run on your computer... Only install plugins from sources you trust."

If you plan to write your firm's rules into the skills, the organization marketplace is also how that private copy reaches advisors. The customization guide covers the sync from a private GitHub repository.

5. Install the Plugin

There are three routes to the same plugin.

1. The plugin browser. In Claude, open Customize in the left sidebar, then Plugins, then Discover or Browse plugins. Find Claude for Financial Advisors and click Add. Anthropic's launch post points Enterprise customers here.
2. The GitHub repository. In the same Plugins area, choose Add marketplace, then Add from a repository, and enter anthropics/claude-for-financial-advisors. Use this route if the plugin does not appear in your browser.
3. The command line. In Claude Code, run claude plugin marketplace add anthropics/claude-for-financial-advisors, then claude plugin install claude-for-financial-advisors.

Plugins work on every paid plan (Pro, Max, Team and Enterprise), per Anthropic's plugin help page. What changes by plan is the admin control and the records tooling around the plugin. Whether each partner connector works outside Enterprise is worth confirming with Anthropic and the vendor, Schwab in particular.

The repository, at version 1.0.0, calls itself a reference implementation, "Not actively maintained or monitored." Treat what you installed as a starting point that your firm keeps current.

6. Run the Onboarding Skill

Have each pilot advisor type /onboarding in Cowork. It is the plugin's guided first run, and it is worth running even for someone who already knows Claude.

It learns about the firm and the advisor's biggest time sinks, then the tech stack across eight categories, from CRM and portfolio reporting to planning and research. It walks through connecting each tool (Settings, Connectors, Connect, then sign in with that vendor's account) and explains how skills work.

Then it offers a demo. Choose the fictional household, the Millers, so no real client data is used on day one. It closes by recommending one skill to try next, based on the time sinks the advisor named.

If an advisor names a system with no connector, onboarding suggests asking that vendor to build an MCP server or setting up a custom one. MCP, the Model Context Protocol, is the open standard Claude uses to talk to other software, and the MCP guide explains what a custom connector involves.

Onboarding also teaches one limit: Claude cannot change its own settings. When an advisor wants a skill to behave differently, Claude writes a short instruction to paste into Settings, General, Instructions for Claude, and that instruction then applies to every conversation the advisor has.

Section 13 explains why firm rules should never live there.

7. Connect Each System, Read-Only First

Connect one system at a time. When something looks wrong in a prep document, you want to know which connection caused it.

Most connectors work at the user level. Each advisor signs in with their own account at the vendor, and the connector inherits that person's permissions in that system. If an advisor cannot see a household in the CRM, Claude cannot see it for them either.

Schwab works at the firm level. Financial Planning reported that the RIA activates the connector for the firm and controls which users can use it, with account numbers, Social Security numbers and clients' dates of birth masked. The plugin's README listed Schwab as coming soon when this guide was written, and until it reaches your firm, the estate and rebalance skills take a paste or upload instead.

Start read-only. The README's security note tells firms to "prefer connectors that expose read-only tools." Where a vendor offers a read-only connection, begin there, and add write access one system at a time, on purpose.

Turn on CRM writes when the pilot starts, since /post-meeting exists to write notes and tasks. Other write access, such as Black Diamond's reports, tasks and rebalance sessions, can wait until after the pilot.

A sensible order: the CRM, then the portfolio platform, since /pre-meeting leans on both. After those come planning and estate (MoneyGuide, Wealth.com), meeting notes (Zocks), alternatives (iCapital) and email.

Any system without a live connector still works by paste or upload. Write that fallback down for each one, so an advisor knows which export to pull when a skill asks. The connectors guide lists what each connector reads and writes.

8. Set Spend Limits and Model Defaults

Enterprise usage is billed at API rates and pooled across the organization, so without a limit, one heavy user draws on everyone's budget. Anthropic lets you set spend limits for the organization and for individual users, and Team has spend controls too.

For a first number, use Anthropic's own ballpark. Peter Nolan, its head of asset and wealth management, put usage at roughly $70 to $120 per user per month for firms that use it actively, TechTimes reported. Set the pilot's per-user limits with that range in mind, then compare against real usage after week one.

Two more choices belong here. If your policies require US-only processing, Anthropic bills US-only inference at 1.1x, so budget for it. And agree which Claude model advisors use for client work, written into the admin handbook, so results from different pilot advisors are comparable.

The first month's bill is the cheapest lesson you will get in how your firm actually uses the tool.

9. Switch On the Compliance API and Connect the Archive

Do this before the pilot touches a real household. A client email drafted during the pilot can be a record like any other.

Three pieces get confused here:

• Audit logs (Enterprise). Owners export them from Organization settings, Data and Privacy, and each export covers the past 180 days of events such as sign-ins, projects, files, and chat creation and deletion. Per the help center, "Title and content of chats and projects are not available to be exported in audit logs."
• The Compliance API (Enterprise, excluding Public Sector). It pulls activity events, chat data and file content, covering Cowork in the Claude apps and Claude Code. It does not cover sessions on Amazon Bedrock or Google Vertex AI, or Claude Code cloud sessions. Only the Primary Owner can enable it, under Organization settings, API (details).
• Your archive. This is the system that keeps records for the five years that Rule 204-2 generally requires. Smarsh Capture for Claude Enterprise, launched in May 2026 on the Compliance API, ingests conversations, prompts, files and activity logs, including deleted and archived conversations and Claude-generated artifacts.

The sequence: the Primary Owner switches on the Compliance API, your archive vendor gets access, and you confirm that a test conversation shows up in the archive. If you use a different archive, ask the vendor whether it captures from the Compliance API, and get the answer in writing.

On Team there is no Compliance API, so your procedure has to move final outputs into the archive by hand. That is a weaker control, and it needs your CCO's written sign-off.

The plugin will not do this part for you. The /compliance skill says plainly, "There is no archiving connector yet," and it tells advisors that its scratchpad file is not the firm's books and records.

10. Write the Policy Addendum

An addendum to your existing AI use policy is enough. It should say:

• Which skills are approved, for which roles, and which connectors each may use.
• Who approves writes to the CRM, and that client-facing drafts go through /compliance and then to the CCO whenever the skill escalates them.
• Which channels client drafts may go out on. The /compliance skill itself warns about off-channel texting, so name the archived, firm-approved channels.
• That client data goes into Claude only inside the firm's workspace, never through a personal account.
• How the CCO supervises: what gets sampled (prep documents, CRM notes, compliance reviews), how often, and where each review is recorded.

Then open a vendor file for Anthropic and one for each connector vendor. Regulation S-P, as amended in 2024, requires oversight of service providers that touch customer information, including notice from them within 72 hours of a breach, and the last compliance date for SEC-registered advisers passed on June 3, 2026.

State-registered advisers answer to their state regulator. Many state rules track the SEC's, so confirm the details with yours.

Treat all of this as operational guidance, with interpretation left to your CCO and counsel. The compliance guide goes rule by rule, and the AI governance audit checklist scores the finished file.

11. Pilot for a Week With Two or Three Advisors

The Millers showed that the plugin runs. The pilot shows whether it works on your households, with your data, in your advisors' hands.

Pick two or three advisors with real client meetings that week, and include one skeptic. Each one runs /pre-meeting before every review, /post-meeting after, and /compliance on every client-facing draft.

Keep one shared log with four columns: what a skill got wrong, what data was missing, what the CCO flagged, and what the advisor would change.

At the end of the week, sort the log:

• Missing data is a connector or permissions fix.
• A wrong format is a customization job for your firm's copy of the skills.
• Anything the CCO flagged goes into the policy addendum or the /compliance rules.

Set the end date before you start, and hold it. A week of real meetings surfaces the problems that matter.

12. Launch

Launch is a settings change and a short meeting.

1. Set the plugin to Required, or Installed by default, for the whole firm in Organization settings, Plugins and skills.
2. Send a one-page version of the policy addendum: what is approved, what goes through /compliance, and who to ask.
3. Name a champion, ideally one of the pilot advisors, to answer questions for the first month.
4. Book a 30-day review of usage, spend and the CCO's samples.

The install is the easy part, and habits take longer. The training guide lays out a four-week plan, and why AI rollouts fail covers what goes wrong after launch day.

13. Common Setup Mistakes

• Connecting everything at once. A dozen connectors on day one means a dozen places a wrong answer could come from, and more access than anyone needs. Connect two systems, prove them, then add.
• Skipping the archive. Audit logs record 180 days of events and no chat content. Until the Compliance API feeds your archive, the conversations sit outside your books and records.
• Letting each advisor put firm rules in their own Instructions. Anything pasted into Settings, General, Instructions for Claude applies to every conversation that advisor has, and binds nobody else at the firm. Firm rules (required disclosures, note format, model names) belong in the firm's copy of the skills, pushed through the organization marketplace, where members cannot edit them.
• Treating a /compliance redraft as approval. The skill calls its output a draft review for the CCO and never calls a redraft compliant. The CCO still decides.

Each of these is cheaper to avoid in week one than to unwind in month three. If you would rather hand the work to someone, the guide to Claude for Financial Advisors consultants compares the options.