Tools

What You'll Build

A research stack that turns every article you've ever read, every earnings call you've taken notes on, every sell-side report you've annotated, every conversation with a fellow investor, into one queryable knowledge base that Claude reads on demand.

Three pieces wired together:

  1. Obsidian holds everything you've ever read or written about the markets, locally, in plain markdown files you own forever.
  2. MCPVault is a small piece of free software that makes the vault readable by Claude.
  3. Claude Code is the chat where you ask questions that span years of your research.

Sample queries the setup lets you answer in 5 seconds instead of an afternoon:

This is what a Bloomberg seat costs $24,000 a year to give you, except yours is built on your own thinking.

Why This Works

The hard part of being a good investor is not reading. It is remembering what you read, six months later, when it matters.

Every analyst has the same workflow problem. You spend hours reading a 10-K, take notes in Notion. You watch an earnings call, drop highlights into a Google Doc. You exchange a thoughtful email with a friend about a position, the email is forgotten in your inbox forever. You annotate a sell-side PDF that lives in your downloads folder until your laptop dies.

By the time the position matters — when the stock drops, when an earnings event approaches, when a competitor's news hits the tape — the research is somewhere. You just cannot find it. You re-read the 10-K. You re-watch the earnings call. The work compounds slowly because every retrieval is starting from scratch.

The fix is two parts. One, get every piece of your research into one local, durable, plain-text vault. Obsidian is the right answer here because the files are yours forever, the format will outlive any subscription, and the tagging is rich. Two, put an AI on top of the vault that can read across years of notes in seconds.

That second part is what changed in the last twelve months. Claude with an MCP connection to Obsidian doesn't just search the vault — it reads it, synthesizes across notes, and gives you the through-line.

An analyst who builds this stack is not just faster. They are working with leverage their younger self did not have. Every note from five years ago is still alive. The mistakes are catalogued. The good calls are documented. The themes the analyst saw before anyone else are searchable by the analyst who saw them.

This is what compounding research actually means.

How the Three Pieces Fit Together

Layer 1: Obsidian as the Vault

Obsidian is a free local-first markdown editor. Every note is a plain .md text file in a folder on your machine. No cloud lock-in. No SaaS that can change pricing on you. No vendor that can delete your archive.

You organize your vault into folders that match how you think about the market. A reasonable starting structure:

/Vault/
  /Companies/
    AAPL.md
    MSFT.md
    NVDA.md
  /Themes/
    AI infrastructure.md
    EV adoption.md
    Healthcare cost compression.md
  /Earnings/
    2026/Q1/AAPL Q1 2026 notes.md
  /Reading/
    2026-05-15 Stratechery on TSMC.md
    2026-05-12 Damodaran NVDA valuation.md
  /Conversations/
    2026-05-10 lunch with Vikram (oil thesis).md

The folders matter less than the frontmatter on each note. Frontmatter is the small block of metadata at the top of every markdown file:

---
ticker: AAPL
event: earnings
date: 2026-04-25
sentiment: bullish
themes: [pricing-power, services-mix, gross-margin]
sources: [10-Q, earnings-call]
---

This metadata is what makes the vault queryable. Every note is tagged. Tags are searchable. The AI can filter on them.

Layer 2: MCPVault Reads the Vault

MCPVault is a free open-source MCP server. You install it once. It runs on your machine alongside Obsidian. It exposes a small set of tools that let Claude:

The point is structured access. A plain "read files" tool would dump 47 raw markdown files into Claude's context and burn through tokens. MCPVault parses the frontmatter, returns structured records, and only sends the parts of notes that match the query. The token efficiency is the difference between this being usable and being unusable.

Layer 3: Claude Code on Top

Claude Code is the chat interface where you do the work. With the MCP connection live, every question you ask is grounded in your actual vault.

The way a session looks:

The whole loop runs inside a single chat. The vault gets richer with every cycle.

Step-by-Step Setup

Step 1: Install Obsidian and Create the Vault

Obsidian is free for personal use. Download it. Create a new vault — call it whatever you want (Research, Markets, Investing). Save the vault folder somewhere durable (a local SSD is fine; iCloud Drive or Dropbox work for cross-device sync but are slower).

Pick a folder structure that matches your thinking. The structure above (Companies / Themes / Earnings / Reading / Conversations) is a reasonable starting point. Don't overthink it. Folders are easy to refactor later because Obsidian rewrites every cross-link automatically.

Step 2: Establish Your Frontmatter Schema

This is the most important decision in the entire setup. Spend an hour deciding what every note will be tagged with. A starting schema:

Save a markdown template in Obsidian's Templates/ folder with this frontmatter block at the top. Now every new note starts with the schema in place.

Step 3: Backfill Two Years of Notes

The setup is only useful if the vault is populated. Plan a weekend to migrate two years of past research into the vault. Order of priority:

  1. Your current portfolio. Every position you hold today, with the most recent thesis note for each.
  2. Earnings notes for your portfolio companies. At least the last four quarters per name.
  3. Themes you actively track. One note per theme, with frontmatter and links to the relevant tickers.
  4. Any conversation or sell-side report that materially shaped a position.

Don't try to backfill everything. The vault gets useful at maybe 50-100 notes. It gets powerful at 300+. Backfill the parts that matter, and let the rest accumulate organically going forward.

Step 4: Install MCPVault

MCPVault is on GitHub at bitbonsai/mcpvault. Installation is a one-line command in your terminal:

npx -y mcpvault@latest --vault /path/to/your/Obsidian/vault

This starts a local MCP server that watches your vault and exposes it to any MCP-compatible client.

Step 5: Wire Claude Code to MCPVault

In Claude Code (or Claude Desktop), add the MCP connection:

claude mcp add mcpvault npx -y mcpvault@latest --vault /path/to/your/vault

The first time you query the MCP, Claude verifies the connection and reads the vault index. Test by asking "how many notes do I have tagged ticker: AAPL?". If it answers with a real number, you're wired.

Step 6: Build Your First Workflow

Pick one stock you cover. Run the full loop on it:

  1. "Summarize my current view on $TICKER from the most recent notes in the vault."
  2. "What's been the consistent bull case across my notes? What's been the bear case? Which side has more recent data?"
  3. "What did I get right in my $TICKER notes from 2024? What did I get wrong? What was I uncertain about that has since been resolved?"

These three questions, asked of your own historical thinking, will produce something you have never had before — a structured view of how your investment process actually performs.

Step 7: The Daily and Quarterly Cycle

Build the habit. Two cadences:

Daily (15-30 min before market open or after close): drop notes from anything you read that day into the vault. Use the template. Tag aggressively. Don't write essays — even three bullet points per article is enough as long as the frontmatter is right.

Quarterly (a half-day at the start of each earnings season): run the full Claude-assisted review on each name in your portfolio. Pre-earnings synthesis. Live notes during the call. Post-call comparison. Synthesis note written back to the vault.

After four quarters of this, the vault becomes your research moat.

Adapting This for Other Information-Heavy Work

The Obsidian + MCPVault + Claude stack works for anyone whose job is to read, remember, and decide.

Corporate strategy. Replace tickers with competitors. Replace earnings with product launches and press cycles. The same pattern produces a competitive intelligence library that compounds.

Venture capital. Tag notes by company, sector, stage, and thesis. Every deal memo, every conversation with a founder, every back-channel reference, lives in the vault. The fund's institutional memory becomes queryable.

Legal practice. Replace tickers with case names. Replace earnings with hearings and filings. A litigator's case archive becomes searchable by argument, judge, opposing counsel, and precedent.

Medical practice. Replace tickers with patient cases (with HIPAA-compliant local-only storage). Themes become symptoms and treatments. The vault becomes a clinical reasoning aide.

Journalism and investigative research. Sources, dates, quotes, documents — all tagged. A reporter's notebook that the reporter can actually query years later.

Academic research. Papers, citations, hypotheses, results. The vault becomes the literature review that updates itself as you read.

The pattern: one local plain-text vault, rich frontmatter, AI on top. The output is the same across every domain. Your past thinking becomes available to your future self.

Gotchas and Tips

What This Replaces

Tool Old cost What it did
Bloomberg Terminal ~$24,000/year Real-time data and historical research
FactSet $10,000+/year Estimates and screening
Sentieo $5,000+/year Earnings transcripts and sell-side research
Junior analyst $60-80K/year Synthesis and research compilation
Your scattered notes unmeasured The work you've already done, mostly lost
What replaces it New cost What you get
Obsidian + MCPVault + Claude $10-30/month A queryable research vault that compounds forever
Quartr (transcripts) $50-100/month optional Earnings transcripts (still useful for raw data)
Your own time, structured same as before But now the time has leverage

The Bloomberg seat is still better at one thing: real-time market data and screening. If you need that, keep it (or use a cheaper substitute like Koyfin at $30/month). What you do not need to pay $24K/year for anymore is somewhere to put your research. That problem is solved.

Keep Reading