Claude Cowork for Accounting Firms: What It Does, What It Costs, and What to Know Before You Connect Anything.

Claude Cowork* is designed to remove a specific friction point that practitioners using Claude daily will recognise. Every new session starts from scratch. You re-explain the context, re-specify the format, and iterate toward something usable. Across a full day of work, that overhead compounds.

Cowork addresses this through two mechanisms: connectors, which give Claude access to data from the tools your firm already uses, and skills, which give Claude pre-built instruction sets for recurring workflows. Together, they shift Claude from a capable assistant you have to brief every time into something closer to a trained team member who already knows the job.

Before going further, there are two practical things firms need to understand: the plan required to access Cowork, and what connecting your firm's data actually means for security and privacy. For the broader picture of where tools like Cowork fit in your practice, see our AI for accounting firms guide.

Last reviewed 22 July 2026 by The Firm's editors.

*The Firm recommends Claude Cowork as a useful tool for accounting and bookkeeping practices. That recommendation comes with a clear qualifier: every firm's circumstances are different. The guidance below reflects our view based on current information, but you should conduct your own due diligence before connecting any client or firm data to this or any AI tool. Where you are uncertain, seek professional advice.

What Plan You Need

Claude Cowork is included on every paid Claude plan. The entry point is Claude Pro, Anthropic's individual subscription at $17 per month billed annually ($200 up front), or $20 if billed monthly. You do not need a Team plan to access the core Cowork functionality, including connectors and skills. Cowork is not available on the free tier.

The Claude Team plan starts at $20 per seat per month billed annually ($25 monthly) for a standard seat, with a premium seat at $100 per seat per month for heavier usage. Team and Enterprise add organisation-wide skill deployment and admin controls suited to multi-user firm rollouts. Importantly for firms, they run under Anthropic's commercial terms rather than its consumer terms (more on why that matters in the security section below). Heavier individual users can also step up to Max, from $100 per month.

Cowork runs on the Claude desktop app for macOS and Windows, and access on the web and mobile is now rolling out in beta. Check current pricing and plan details at Claude pricing before committing, as Anthropic's plan structure continues to evolve. If you are weighing Claude against Microsoft's offering, our Copilot vs Claude comparison for accountants covers the trade-offs.

What Connectors Do

A connector links Claude Cowork to a platform your firm already uses. Once connected, Claude can read and retrieve data from that platform directly inside a conversation. No copying, no manual uploads.

The connector directory has expanded considerably since Cowork launched. The connectors with a clear accounting application include Gmail, Google Drive, Google Calendar, Notion, Slack, Asana, Monday.com, HubSpot, Box, Atlassian, and Linear. Anthropic has since added Google Workspace and DocuSign connectors, plus Microsoft 365 add-ins that let Claude work directly inside Excel, PowerPoint, and Word, with Outlook to follow.

In practice, that means Claude can read a client email thread before drafting a response, pull figures from a workpaper stored in Drive, check what jobs are overdue in Asana, or reference your firm's SOPs in Notion before completing a task. The data is already there. The connector makes it accessible.

What Skills Do

A connector gives Claude access to data. A skill tells Claude what to do with it.

Skills are pre-built instruction sets installed into Claude Cowork and called on demand using the /skills command. Instead of writing a prompt from scratch each time you need Claude to complete a recurring task, the skill contains the context, structure, and logic Claude needs to execute immediately. You load the skill, Claude understands the workflow, and the task gets done.

The practical impact is real. Less manual prompting per session. Fewer tokens used across repeated tasks. More consistent outputs, because the logic is baked in rather than reconstructed from memory each time.

How to Build One

A skill is a markdown file. No coding required.

You write three things: the context Claude needs, the steps to follow, and the format you want the output in. Save it, install it into Claude Cowork, and call it with /skills whenever you need it.

The description field is worth getting right. It tells Claude when to load the skill automatically, so it can trigger without you calling it manually every time.

The best starting point is a task you already prompt Claude for repeatedly. Write down the steps, add the context Claude needs, specify the output format, and you have a skill. You no longer have to start from a blank page either: Anthropic now ships ready-made agent templates, including a set built for finance workflows, and we have collected the ones worth stealing in our guide to AI agent templates for accounting firms.

The Workflows Worth Building Skills For

These are the accounting tasks where skills produce the clearest return:

• Month-end close agent (we walk through building one in our month-end close with Claude guide)
• Tax workpaper populator
• Budget vs. actuals reporter
• Missing documents tracker
• Bank reconciliation builder
• Invoice and receipt extractor
• Depreciation scheduler
• ATO correspondence interpreter
• Client email drafter
• Meeting prep summariser
• Proposal drafter
• Workflow status report

Where It Gets Useful

The real value comes when connectors and skills work together. A meeting prep skill paired with Gmail and Google Calendar means Claude reads recent client correspondence, checks what is scheduled, and produces a briefing note in one step. A missing documents skill paired with Google Drive means Claude scans the file, identifies what is absent, and drafts the follow-up without you touching it.

Cowork has also grown more autonomous since it launched. Scheduled tasks can now run in the cloud on a recurring basis (a Monday-morning WIP report or a daily inbox triage that runs without your machine being online), and Cowork can split larger jobs across parallel sub-agents, so a month-end pack and a client follow-up list can be built at the same time. Cowork asks permission before taking actions by default, and you control how much autonomy it gets.

Each skill removes a recurring prompt. Each connector removes a manual step. Together, they shift Claude from something you manage constantly into something that mostly runs in the background.

For firms already using Claude daily, this is the logical next step. Start with the task you prompt Claude for most often, write the skill once, and you will not have to write that prompt again.

The benefit compounds quickly:

• Time previously spent re-prompting Claude gets redirected to work that actually needs your judgement.
• Output quality stays consistent across your team regardless of who runs the skill; and
• A single practitioner can carry a heavier workload without the administrative overhead that usually comes with it.

Security and Data Handling: What Firms Need to Know

This is the section most practitioners will want to read carefully. Accounting and bookkeeping firms handle some of the most sensitive data in their clients' lives. Connecting that data to any AI tool requires a clear-eyed view of what is actually happening.

Understand the data flow before you connect anything

When you link a connector to Gmail or Google Drive, Claude Cowork can read the contents of emails and documents within those platforms. That access is intentional and is what makes the tool useful. It also means Claude is processing data that may include client financial records, correspondence, and personal information. Before enabling a connector, understand what is being accessed, how it is handled, and whether that is appropriate for your firm's obligations.

Our recommendation: Enable connectors one at a time. Start with a single low-risk platform, such as your internal Notion workspace or Google Calendar, before connecting anything that holds client data. Confirm it is working as expected before expanding access.

Check your model-training settings: the defaults matter

This is the area that has changed most since Cowork launched. Under Anthropic's current consumer terms, conversations on the Free, Pro, and Max plans can be used to train Claude's models unless you opt out in your Privacy Settings, with retained data held for up to five years. The Team and Enterprise plans sit under Anthropic's commercial terms, which do not use your data for training by default. In other words, simply being on a paid plan is no longer enough: the plan type and your settings both matter.

Our recommendation: If you use Claude Pro or Max for any client-related work, turn model training off in Privacy Settings before you connect anything, and confirm the setting applies across your firm's accounts. For multi-user firms, a Team or Enterprise plan under commercial terms is the cleaner answer. Verify the current data handling terms directly at anthropic.com/privacy and anthropic.com/usage-policy, as policies can change. Do not rely on free-tier Claude for sensitive professional work.

Review your obligations under Australian privacy law

Accounting firms collecting, storing, and processing client personal and financial information are subject to the Privacy Act 1988 and the Australian Privacy Principles. Before connecting client data to any third-party tool, satisfy yourself that doing so is consistent with your privacy obligations, your engagement letters, and any confidentiality undertakings you have given to clients.

Our recommendation: Before connecting any client-facing data, review your privacy policy and engagement letter templates to confirm they cover the use of AI tools in your workflow. If they do not, update them. Where you are uncertain about your obligations, seek legal advice. This is one area where caution now prevents significant exposure later.

Start with internal data, not client data

A meeting prep skill pulling from your own firm's Notion SOPs and Google Calendar carries lower risk than a connector reading live client email threads. The connectors offer more value when used thoughtfully.

Our recommendation: Run Cowork on internal firm data for at least four weeks before connecting anything that touches client information. Use that period to understand what Claude accesses, what it produces, and where human review is still needed. Build the habit of reviewing outputs before acting on them.

The security of connected platforms flows directly into Cowork

Claude has no persistent memory across conversations unless memory features are explicitly enabled. Each session is discrete. This limits certain risks, but it also means the security of the underlying platforms you connect matters significantly. The strength of your Google or Microsoft account security becomes the baseline protection for anything Claude can access through a connector.

Our recommendation: Before connecting any platform, confirm that multi-factor authentication is enabled on that account across your firm. This is basic hygiene, but it is the most direct control you have over what Claude can access through a connector.

Treat outputs as a strong first draft

Claude is capable but not infallible. A reconciliation built by a skill still needs a review. A workpaper populated by Claude still needs sign-off. The skill removes the manual effort. The professional judgement remains yours.

Our recommendation: Build a review step into every skill you create. Before deploying a skill across your team, run it on known data and check the outputs against what you would expect. If it fails on edge cases, refine the skill before it runs unsupervised.

The Logical Next Step for Firms Already Using Claude

For firms already prompting Claude daily, Cowork is a meaningful step forward. The productivity gains are real. The workflow implications are significant. So is the responsibility that comes with connecting firm and client data to any AI tool.

Start with the task you prompt Claude for most often. Write the skill once. You will not have to write that prompt again.

And before you connect anything to client data, make sure you understand exactly what you are connecting, and why.

Get started with Claude Cowork

Frequently asked questions

What does Claude Cowork cost?

Cowork is not sold separately. It is included on every paid Claude plan. Claude Pro is $17 per month billed annually ($200 up front) or $20 billed monthly. Team seats start at $20 per seat per month billed annually, and Max plans for heavy individual use start at $100 per month. It is not available on the free tier. Check Claude pricing for current figures.

Is Claude Cowork safe for client data?

It can be, with the right setup, but not by default. On the consumer plans (Free, Pro, Max), conversations can be used for model training unless you opt out in Privacy Settings. Team and Enterprise plans run under commercial terms that do not train on your data by default. Beyond that, safety depends on your own controls: MFA on connected accounts, connectors enabled one at a time, engagement letters that cover AI use, and human review of outputs. Verify the current terms at anthropic.com/privacy before connecting anything.

What do accounting firms actually use Cowork for?

The highest-return uses we see are recurring, structured tasks: month-end close checklists, tax workpaper population, budget vs. actuals reporting, missing-document chasing, bank reconciliation prep, meeting briefings drawn from email and calendar, and client email drafting. Our month-end close with Claude guide walks through a full example.

How is Cowork different from regular Claude chat?

Chat starts every session from a blank slate: you brief Claude each time. Cowork adds connectors (direct access to tools like Gmail, Drive, and Excel), skills (saved instruction sets for recurring workflows), scheduled tasks that run without your device online, and the ability to work on files and multi-step jobs autonomously, asking permission before it acts. Same underlying model, far less repeated setup.

How do we get started with Cowork?

Get on a paid plan, check your privacy and training settings, then start small: connect one low-risk internal tool (a calendar or internal wiki, not client email), and build one skill for the task you prompt Claude for most often. Run it on internal data for a few weeks with a review step before touching client information. Our AI for accounting firms guide covers the wider rollout playbook.