Claude Finance Agents

I invest in a lot of stocks. I'm also terrible at managing them.

Not the picking part. The maintenance part. Reading earnings transcripts. Updating models. Tracking guidance vs actuals. Knowing when management's story has quietly drifted from the thesis I bought into 18 months ago. That work is what professional analysts do all day, and it's the work I never get to because I'm running businesses.

For years my portfolio review process was: check the price, feel something about it, move on. Anyone who runs operations and invests on the side knows this gap. You're great at the business stuff. You're chronically behind on the portfolio stuff. And hiring someone to do it for you doesn't make sense at retail size.

Anthropic just shipped a marketplace of agents that closes the gap. Three of them are built for retail investors and founders managing their own portfolios. The other seven are for sell-side analysts and fund admin teams.

What it actually feels like now: I get daily, weekly, or monthly reports on every name I own, written by Claude, at the level of a junior analyst's first three months on the job. I never have to read a 40-page transcript again. The work that used to be a week of evenings is now one chat per stock per quarter.

This is the install. Plus the prompts I actually run. Plus the Microsoft 365 upgrade that makes it 10x more useful.

What you're installing

Three agents from the Anthropic Financial Services marketplace. Free with any paid Claude plan ($20/month Pro and up):

• Earnings Reviewer. Drop in any earnings call and filings. Get a model update and a draft research note on the stock you own.
• Market Researcher. Pick a sector or theme. Get an industry overview, competitive landscape, peer comps, and a shortlist of names worth deeper analysis.
• Model Builder. Name any company. Get a full DCF, LBO, 3-statement model, and comps live in Excel.

This is the analyst desk Anthropic just gave you for free.

The 4-step install

1. Install Claude Cowork. Download Claude Desktop at claude.ai/download. Open it, switch to the Cowork tab. Cowork is the desktop mode where plugins run. Web Claude doesn't have plugins.
2. Add the Financial Services marketplace. Settings → Plugins → Add plugin. Paste:

https://github.com/anthropics/claude-for-financial-services

Hit Add. The marketplace loads with all 10 agents and supporting plugins.

1. Install the 3 retail-relevant agents:

• earnings-reviewer
• market-researcher
• model-builder

Each agent plugin is self-contained. Installing brings the skills it needs with it.

1. Optional: install the financial-analysis core. Adds standalone slash commands (/comps, /dcf, /earnings) outside the agents, plus 11 institutional data connectors (Daloopa, Morningstar, S&P Global, FactSet, Moody's, MT Newswires, Aiera, LSEG, PitchBook, Chronograph, Egnyte). Most require their own subscription. Skip unless you already pay for institutional data.

Five minutes. Done.

Earnings Reviewer

Use it for: post-earnings reaction on stocks you own, pre-earnings prep on your watch list, catching up on a quarter you missed.

First-run prompt:

"I own [TICKER]. Here's the latest earnings call transcript [paste or attach PDF] and the 10-Q filing. Update my model with the new revenue, margin, and guidance numbers. Flag anything that materially changes my thesis on the stock and draft a research note I can save to my portfolio file."

What comes back: an updated Excel model with the new quarter's actuals filled in, plus a draft research note covering the key numbers vs consensus, the 3-5 most important things management said on the call, and a thesis-impact flag (still holds / needs review / broken).

Follow-up that finds the pattern:

"Compare the last 4 quarters of [TICKER] guidance to actual results. Show me where management has been over-promising or under-promising. What does that pattern mean for next quarter's setup?"

Common mistake: running it without pasting in the actual transcript. Without the live document, the agent works from training-data context that's months old. Always feed it the current call.

Why this saves time: 60 to 90 minutes per stock per quarter compresses to about 5 minutes. Ten stocks owned = 10+ hours saved every earnings season.

Market Researcher

Use it for: sector or theme research before committing to new positions. Industry analysis. Finding new ideas adjacent to what you already own.

First-run prompt:

"I'm researching the [SECTOR or THEME, e.g., 'GLP-1 weight loss drugs', 'AI semiconductor supply chain', 'commercial real estate REITs']. Give me an industry overview, competitive landscape, peer comparison across the major players, and a shortlist of 5 to 10 stocks worth deeper analysis."

What comes back: structured sector report covering the major players, market dynamics, competitive moats, and a screened ideas list ranked by relevance to your criteria.

Follow-up that surfaces the asymmetry:

"Take the top 3 names from your shortlist. Run a side-by-side comparison: revenue growth, margin trajectory, valuation multiples, key risks. Tell me which one has the most asymmetric upside if my thesis on [SECTOR] plays out."

Common mistake: treating it like a news aggregator for stocks you already own. Market Researcher is for new-idea generation, not portfolio monitoring. For news on existing holdings, use Earnings Reviewer when they report.

Why this saves time: a sector deep-dive that takes hours of broker reports, Substack posts, and Wikipedia tabs collapses to one chat. Especially valuable when entering a vertical you don't already follow.

Model Builder

Use it for: when you're seriously considering a position and need to do the math before committing capital. Stress-testing a stock you already own. Comparing companies head-to-head.

First-run prompt:

"Build me a full 3-statement model and DCF for [TICKER]. Pull the last 4 years of filings. Project 5 years forward. Show me a fair value range with sensitivity analysis on revenue growth and margins. Highlight the 2 to 3 assumptions the valuation is most sensitive to."

What comes back: an editable Excel file with income statement, balance sheet, cash flow, and a DCF tab. All assumptions labeled and color-coded. Output is structured the way an investment banking analyst would build it in their first three months on the job. Model Builder also handles LBO and comparable company analysis if you ask.

Follow-up that pressure-tests the thesis:

"Run the same DCF with three scenarios: bear case (revenue grows half as fast as guidance), base case (matches guidance), bull case (10% above guidance). Show me what stock price each scenario implies and where my margin of safety is."

Common mistake, the dangerous one: trusting the output without sanity-checking the assumptions. A clean model with bad assumptions gives you a clean, professional, completely useless valuation. Always verify the WACC, perpetual growth rate, and revenue trajectory before relying on the fair value.

Why this saves time: modeling work that takes a PE associate 1-2 days compresses to one chat plus 30 minutes of assumption review.

How I actually run this. The briefing workflow

This is the part nobody else talks about. The three agents are powerful on their own. Stacked together with a routine, they become the equivalent of having a junior analyst on retainer.

What I run, on what cadence:

Daily (5 minutes, only on earnings days): If anything in my portfolio reported, I drop the transcript into Earnings Reviewer with the standard prompt. Comes back with the thesis-impact flag. If it's "still holds," I move on. If it's "needs review" or "broken," I read the full note.

Weekly (15 minutes, Sunday night): I ask Claude to summarize anything that's moved in my watchlist over the last 7 days. Price action, news, analyst updates. Not for trading decisions. For staying current on names I'm watching but haven't bought yet.

Monthly (30 minutes, first weekend of the month): I run Earnings Reviewer's pattern-comparison follow-up across every name I own. The one that asks where management has been over- or under-promising. That single prompt has surfaced more thesis breaks than my old "read everything religiously" approach ever did, because it catches the pattern across quarters that a single-quarter read misses.

Quarterly (one afternoon): Full Model Builder rebuild on the names I'm most invested in. Fresh DCF, fresh assumptions, fresh fair value. This is the discipline check. Am I still in the trade for the reason I bought in?

That's the whole system. Running a real business, getting the financial workflow of a fund manager, on a $20 plan, in less time per week than most people spend scrolling Twitter for stock takes.

The Microsoft 365 upgrade

If you live in Excel, PowerPoint, or Word for portfolio work, this is bigger than the agents themselves.

As of April 2026, Claude runs directly inside Excel, PowerPoint, and Word as native add-ins. Outlook is in public beta. Context carries between apps automatically.

What this looks like in practice:

• In Excel. Open a model Model Builder generated. Ask Claude to audit the formulas, run sensitivity analysis, or pull in fresh data from a filing.
• In PowerPoint. Turn that model into a stock pitch deck where slides update when underlying numbers change.
• In Word. Draft an investment memo or quarterly review that pulls structure from your model and language from your notes.

Install at claude.com/claude-for-microsoft-365. Included with any paid Claude plan, no extra charge.

The other 7 agents

The marketplace also has Pitch Agent, Meeting Prep Agent, Valuation Reviewer, GL Reconciler, Month-End Closer, Statement Auditor, and KYC Screener. Excellent if you sit in a sell-side coverage, fund admin, or compliance seat. Background context for everyone else. Don't install them unless your job is one of those.

What this changes

The asymmetry used to be access. The buy-side desk had Bloomberg, Capital IQ, FactSet, and a team of analysts running models 18 hours a day. The retail investor had Yahoo Finance and a free transcript on Seeking Alpha if they were lucky.

That asymmetry is closing. Not because retail tools got better. Because the same analyst work that used to require a team now runs on a $20 plan and one good prompt.

The people who own their own portfolios and learn to operate this stack are about to look very different from the people who don't. Not because they pick better stocks. Because they finally have the time and infrastructure to actually manage what they already own.