5 Ways to Use Claude Cowork for Corporate Treasury

Claude Cowork became generally available in April 2026, and within weeks treasury teams started considering how and where it could fit into their workflow.

If you’ve missed it, Cowork is Anthropic's agentic AI for knowledge work. Unlike a chat window, it operates on your desktop with controlled access to local files, applications, and connectors, executing multi-step tasks rather than answering one prompt at a time. It's the agentic AI architecture that powers Claude Code, repackaged for non-developers and wrapped around outcomes instead of conversations.

For corporate treasury, the appeal is immediate. Most days, the function runs on a steady cadence of repeatable work, including pulling balances, reconciling positions, drafting briefings, chasing maturities and prepping the morning narrative for the CFO. It’s not necessarily technically complex, but all of it is time-consuming. Cowork is built for exactly this category of task.

But before the function pivots wholesale to a new desktop agent, there are questions worth answering. What can Cowork actually do for a treasurer today? Where are the data, security, and governance limits? And how does it fit alongside the AI capabilities already embedded inside a modern treasury platform?

This is a practitioner's view of how to put Claude Cowork to work in corporate treasury, where it complements rather than replaces purpose-built treasury AI, and what's coming next as Anthropic and treasury technology vendors converge through APIs and the Model Context Protocol.

What Claude Cowork Actually Does

Cowork sits as a tab inside the Claude desktop app, alongside Chat and Code. You describe an outcome, attach the relevant folders or connectors, and Cowork plans the steps, executes them, and surfaces the result for review. It can read and write local files, open browser sessions through Claude in Chrome, schedule recurring tasks, and produce polished deliverables like formatted spreadsheets, slide decks, briefing documents,  without the user shepherding each step.

Cowork runs in a sandboxed virtualization layer that limits its filesystem access to folders the user explicitly grants. Enterprise admins can manage feature access, control egress, monitor activity through OpenTelemetry, and govern which connectors are available to the organization. Anthropic has been clear that Cowork activity is not currently captured in audit logs or the Compliance API, which is an important point to note. Cowork is not the right place for regulated workloads where a full audit trail is mandatory.

That last point shapes the practical answer to what should treasury actually do with it.

Five Treasury Workflows Where Cowork Earns Its Seat

The work that suits Cowork is high-effort, repeatable, and sits one or two steps removed from the system of record. Treasury has plenty of it.

The Morning Briefing

Most teams produce some version of a daily liquidity snapshot of total cash, investments, short-term debt, net position, anything overdue and anything maturing in the next 48 hours. Cowork can pull the underlying numbers from a treasury platform export, layer in market context (rates, FX moves, FOMC commentary), draft the email, and queue it in Gmail or Outlook. With scheduled tasks, the same workflow runs every weekday at 6:30 a.m. without anyone touching it.

Variance Commentary

When a forecast versus actual variance lands above threshold, someone has to explain it. Cowork can pull the underlying transaction data, surface the largest contributors by entity or counterparty, draft a narrative, and flag items that need human follow-up. The treasurer reviews and signs off rather than starting from a blank cell.

Counterparty and Bank Fee Reviews

Periodic reviews, analyzing fees across banks, comparing money market yields against IORB, summarizing counterparty exposure changes, are well-suited to a Cowork session. Point it at the relevant statements or exports, give it the framework, and get a structured comparison back.

Policy and Audit Prep

Cowork can scan a folder of internal policy documents, pull the relevant clauses for an upcoming audit, and produce a working draft of responses to auditor questions. It won't replace the controller, but it eliminates the first three hours of document gathering.

Recurring Board and Management Reporting

The KPI deck, the rolling 13-week update, the quarterly liquidity narrative, all of these have a stable structure and changing inputs. Cowork can refresh them on a schedule, pulling the latest data and producing a draft for the treasurer to refine.

Each of these has the same shape: structured inputs, a clear output format, and a human review step before anything goes external. That's the Cowork sweet spot.

How MCP Brings Everything Together

The Model Context Protocol (MCP) is the standard that connects AI models to external tools and data sources. It's the substrate that makes agentic AI useful, allowing the model to query approved data sources and call approved tools directly, with permissions enforced at the connection layer, rather than the model reasoning over text someone pasted into a prompt.

Trovata AI 2.0, released in March 2026, is built on MCP as foundational infrastructure. That's what makes Trovata AI Agents able to work on real treasury objects — pulling live data from the platform, running on schedules or triggers, and producing grounded answers rather than plausible-sounding but unreliable output. Together with Chat and Insights, Trovata AI Agents are how treasury teams put AI to work inside the platform, where governance and depth of context matter most.

That data foundation isn't limited to in-platform work. A treasury team can point Claude Cowork at Trovata's developer APIs today and have the desktop agent query treasury data, run analyses, and orchestrate work across both the platform and the desktop environment.

A more packaged MCP experience from Trovata is on the near horizon, but the underlying connection is already live for customers willing to wire it up through Trovata's developer API. The Trovata team has been working with early customers on this exact pattern, and the trajectory is clear. As more treasury platforms expose MCP-compatible interfaces, the line between 'AI inside the platform' and 'AI on the desktop' becomes a routing decision rather than an architectural one.

Meaningful options for connecting Claude directly to treasury data through MCP are landing imminently., with the same controls treasury teams expect from any production integration.

For teams ready to put this to work, the focus is less on Cowork itself and more on the data foundation. An MCP-ready treasury platform — normalized data, real-time bank connectivity, and a developer-grade API surface — is the prerequisite. Without it, no amount of agentic AI on the desktop will produce reliable output.

A Practical Approach To Getting Started

For a treasury team curious about Cowork, the right starting point is small, internal, and review-heavy. Three suggestions:

1. Start with the morning briefing. Pick the daily report that takes someone 30-45 minutes every morning. Build a Cowork task that produces a draft from the data you already export. Run it in parallel with the human-produced version for two weeks, compare, then adjust.
   

2. Use Cowork on documents, not data. The fastest wins are formatting, summarizing, drafting, and synthesizing. For example, taking five PDFs of bank fee schedules and producing a comparison table, taking last quarter's board narrative and updating it for this quarter and taking a 40-page audit request list and producing a triaged response plan. These are the workflows where Cowork's agentic capabilities show up most clearly.
   

3. Keep the system-of-record data inside the system of record. For anything involving real bank balances, payment instructions, or counterparty exposures, lean on the AI capabilities embedded inside the treasury platform, where governance is built in. The team at CSG has been thinking publicly about this division of labor, and the pattern they describe, treasury becoming more about system design than day-to-day execution, maps directly onto how Cowork and platform-native AI fit together.

The Bottom Line

Claude Cowork is a meaningful capability for corporate treasury, and the teams experimenting with it now will have a working understanding of agentic AI well before it becomes table stakes. The mistake is treating it as a replacement for purpose-built treasury technology rather than a complement to it.

The work that benefits most from Cowork sits in the spaces between systems, such as the briefings, the summaries, the document workflows and the recurring deliverables that humans produce on top of treasury data. The work that benefits most from in-platform AI sits inside the system of record itself, where governance, normalization, and grounded context matter.

Trovata AI Agents handle the core treasury work inside the platform. The Trovata MCP server lets that same data foundation reach Cowork and any other leading AI environment when teams need it. That's the full picture — and it's where treasury technology is going.

See how Trovata AI puts agentic intelligence directly inside your treasury workflows, with the security, governance, and grounded data foundation that desktop AI tools alone can't match. Book a demo to see Trovata AI Agents in action.