A self-contained repo that turns your Claude Code into an equity-research agent. Unlocks Anthropic's official equity-research skills for every investor — a single data MCP that replaces a $150K-per-year institutional data stack, and a personalization layer that caters to sell-side analysts, portfolio managers, retail investors, and academic economists alike.
Three components make it work:
-
Skill library. 24 analysis workflows — Anthropic's official Apache-licensed equity-research bundle (9 skills) plus a community-maintained library (15 skills). Covers idea discovery, single-company deep dives, position monitoring, and macro research.
-
Accessible interface. Claude speaks to you according to your financial proficiency and preferences — adapting tone, depth, and jargon density to match a sell-side analyst, a portfolio manager, a retail investor, or an academic economist. Four unified slash commands —
/discover,/analyze,/monitor,/macro— auto-route plain-language requests across all 24 skills. -
Single data MCP.
drillrconsolidates everything the skills need: structured financial metrics, SEC filings, earnings-call transcripts, company ontology (suppliers / customers / competitors), alternative data (federal contracts, hiring, patents, trade flows), and macro / market context. Covers US and Japanese equities. Generous free quota for individual users.
Anthropic recently open-sourced an excellent equity-research skill bundle — nine institutional workflow templates (initiation note, earnings analysis, catalyst calendar, morning note, thesis tracker, and more). They significantly raise Claude Code's ceiling for equity work. Two gaps remain.
-
Expensive data connector. The Anthropic skills are pure methodology — no data ships with them, so you have to wire in your own connectors. Anthropic's official reference repo wires the skills to eleven institutional MCPs — FactSet, LSEG, S&P Global, Morningstar, Moody's, PitchBook, and others — typically $15–30K per seat per year each, well over $150K combined. Out of reach for independent analysts, academic economists, and anyone running serious workflows without an institutional budget.
-
Targets sell-side professionals, no accessibility layer. The bundle's templates are built around the daily output of an equity research desk — initiation notes, morning notes, earnings previews. The concepts are universal; the jargons (A/E year notation, basis-point shorthand, sell-side report shapes) read arcane to retail investors and to academics doing exploratory company-level work. Out of the box, the agent doesn't adapt to the user's financial fluency or preferred tone.
This project addresses both. The drillr MCP consolidates the eleven-MCP stack into a single connector with a generous free quota; a memory-driven interface layer adapts the output register to whoever is at the keyboard.
You need Claude Code installed.
git clone https://github.com/prof-little-bear/cc-equity-research.git
cd cc-equity-research
claudeInside Claude Code, run /mcp to check the drillr connection and authenticate if prompted (the repo's .mcp.json declares the server — Claude Code picks it up on launch). After that you're ready — type what you want, or fire one of the four slash commands.
Each command opens a short menu of lenses in that category. Pick one by name, or just describe what you want and the dispatcher routes from there.
| Command | Category | What it covers |
|---|---|---|
/discover |
Idea generation | themes, supply chain, alt-plays, federal contracts; plus Anthropic's idea-generation (systematic screens) and sector-overview |
/analyze |
Single-company deep work | business model, earnings scorecard, forensics, reporting drift, management; plus Anthropic's initiating-coverage, earnings-preview, earnings-analysis, model-update |
/monitor |
Position tracking | watchlist, thesis check, event radar; plus Anthropic's thesis-tracker, catalyst-calendar, morning-note |
/macro |
Economic research | yield curve, trade flows, labor market |
Plain language works too. "Run forensics on NKE", "what's coming up for PLTR over the next 6 weeks", "is the labor market softening" — Claude routes through CLAUDE.md's intent map directly to the skill. The slash commands are the discoverability layer; plain language is the power-user layer. Same skills underneath.
Examples: /analyze NKE forensics → runs financial forensics on Nike. /macro → opens the macro menu. /discover what's working in AI infra → routes to themes.
The slash surface stays at four commands no matter how many skills get added — new skills are added inside the dispatchers, not as new commands.
Two project-local files shape every response. This is the layer that addresses Problem #2 — same analytical rigor, different register depending on who's at the keyboard.
.claude/mode.md—new(default) shows orientation on session start;experiencedskips it. Switch by editing the file or telling Claude "I'm experienced now.".claude/style.md— four fields control how Claude communicates:experience(experienced / intermediate / learning),depth(quick / balanced / deep),tone(professional / institutional / conversational / educational), and optionalcoverage(sector focus). Defaults are sophisticated-but-accessible.
Claude reads both at session start, applies them every turn, and absorbs preference shifts mid-conversation. Start using A/E notation casually and tone promotes to institutional. Ask what a term means and experience shifts to intermediate. The file updates and the change is confirmed in one line.
Note: these files are project-local — they live inside the repo, not in Claude Code's cross-session auto-memory. Cloning to a different machine means starting fresh unless you sync the repo.
Anthropic bundle (anthropic-equity-research-skills/) — nine workflow templates vendored from anthropics/financial-services (Apache 2.0): initiating-coverage, earnings-preview, earnings-analysis, model-update, morning-note, catalyst-calendar, thesis-tracker, idea-generation, sector-overview.
Community extension (community-skills/) — fifteen analyst-contributed lenses across four areas: discover/ (themes, supply-chain, alt-plays, gov-contracts), analyze/ (business-model, earnings-scorecard, financial-forensics, reporting-quality, management), monitor/ (watchlist, thesis-check, event-radar), economic-research/ (yield-curve, trade-flows, labor-market).
Each skill is a short markdown file — read one to see exactly what it does.
Data — the drillr MCP
One MCP backs every skill. Six data domains:
- Structured fundamentals — statements, 60+ standardized metrics, consensus estimates
- SEC filings — 10-K, 10-Q, 8-K, proxy, S-1, S-4 with full-text search
- Earnings calls — transcripts and structured summaries
- Company ontology — suppliers, customers, competitors, peer groups, founder backgrounds
- Alternative data — federal contracts, hiring, web/app metrics, patents, trade flows, insider transactions
- Macro and market context — rates, credit, labor, sentiment, indices, commodities, FX, crypto
Sourced directly from primary sources (SEC EDGAR, company IR pages, government databases, customs filings, public market venues) via agentic AI — not resold proprietary feeds. Generous free quota for individual users; no FactSet, LSEG, S&P Global, or Morningstar subscription required.
Coverage: US equities, Japan equities, and ADRs. You don't write SQL — describe what you need, and Claude pulls the data when a skill runs.
Community-skill contributions are very welcome — they're the part of the toolkit that gets sharper as working analysts share what they actually do.
Three small edits for any new community skill:
- Add the skill file itself in
community-skills/<area>/ - Add a one-line entry to the capability map in
CLAUDE.md - Add a row to the matching dispatcher's menu in
.claude/commands/<area>.md
Then submit a PR. The Anthropic bundle is vendored upstream — suggested changes there belong in anthropics/financial-services, not in this repo.
See CONTRIBUTING.md for the skill template, what good skills look like, and review expectations.
The toolkit (community skills, scaffolding, dispatchers, documentation) is licensed under Apache 2.0 — see the top-level LICENSE file. The vendored Anthropic equity-research bundle is also Apache 2.0; see anthropic-equity-research-skills/NOTICE.md for attribution and the upstream-sync command.
Used in this project:
- anthropics/financial-services — Anthropic's open-source equity-research skill bundle, vendored at
anthropic-equity-research-skills/(Apache 2.0) - Drillr — the single data MCP backing every skill (fundamentals, SEC filings, company ontology, alt-data, macro / market signals)