Skip navigation
this github repo from anthropic feels like CHEATING in trading
0:00 / 0:40

this github repo from anthropic feels like CHEATING in trading

Сarm1ne

@carm1nee18K followers

Visit on X
706 likesDownloadOriginal post

48,920 views

this github repo from anthropic feels like CHEATING in trading banks pay analysts six figures to pull comps, build DCFs and rebuild pitch decks at 2am - Anthropic just put 10 AI agents that do that work on GitHub, and anyone can install them in Claude Code it's called Claude for Financial Services. official Anthropic repo, 40.7k stars, Apache-2.0 first understand how it works: > Pitch Agent: comps, precedents, LBO → a branded pitch deck, end to end > Market Researcher: sector → competitive landscape, peer comps, ideas shortlist > Earnings Reviewer: earnings call + filings → model update → note draft > Model Builder: DCF, LBO, 3-statement, comps - live in Excel > + KYC Screener, GL Reconciler, Month-End Closer, Statement Auditor... 10 agents total > they plug into FactSet, S&P Capital IQ, Morningstar, LSEG and PitchBook through MCP here is how to run it yourself: 1. open Claude Code and add the marketplace: claude plugin marketplace add anthropics/financial-services 2. install the core first - comps, DCF, LBO, 3-statement + the data connectors: claude plugin install financial-analysis@claude-for-financial-services 3. add the agents you want: claude plugin install market-researcher@claude-for-financial-services claude plugin install pitch-agent@claude-for-financial-services 4. type /comps NVDA and watch it pull the peer set and build the table 5. type /dcf NVDA - the model lands in Excel with a WACC x growth sensitivity grid 6. ask Claude to run a Monte Carlo on the valuation. you want a range, not one number 7. /ic-memo turns the whole thing into an investment committee memo 8. everything comes out as a DRAFT for human review. check every number before it goes anywhere i won't say anything more:

11 Comments

Comments from the original post on X (Twitter)

  • bodila's profile picture

    @51bodila

    damn mate! repo from Anthropic is actually feels alpha

    1 like

  • 456X's profile picture

    @OptionKing666

    Having 10 dedicated agents plug directly into FactSet and PitchBook via MCP makes automating routine models practical. Just ensure every generated valuation is treated strictly as an unverified draft needing manual human review.

    1 like

  • chuplung's profile picture

    @choopyplug1

    this is seriously next level. the fact that all these agents can work together with real financial data is wild

  • Guilherme Alves's profile picture

    @souguialves

    pulling comps and building the dcf is the easy half. someone still has to own the growth and wacc assumptions, so the 2am job turns into reviewing the agent's model instead of building it

  • COSMOS's profile picture

    @outofcosmos

    trading

  • Jaime Garcia's profile picture

    @JaimeGarci61246

    Pero pone la repo desodorante de bolita

  • Gian Maurice's profile picture

    @gianmauric

    the data connectors are the real gate here, since FactSet and Capital IQ need paid licenses. without those, does the Model Builder just pull from filings and public numbers?

  • 安叫兽|Bird🕊️ 🔶 BNB's profile picture

    @ajs6888

    把投行分析师的活拆成十个 agent,这步走得也太快了。

  • 抓一把星星's profile picture

    @imscarletD

    Great point!

  • catman's profile picture

    @catmanyau

    The draft-only output and human review are the strongest safeguards here. I wouldn't call the workflow itself cheating: generating a DCF is the easy part; validating assumptions and data is still the hard part.

  • Harry Read's profile picture

    @HarryRead33

    This is good but the issue is that it starts from zero context, will go away for 20 mins and then pull you the best that it managed. If you extract all the relevant info once, and feed pre-computed outputs with all of the peers you can get much better outputs. lmk if you want to swap notes