Best AI Agent Skills for Data Analysts in 2026

For full setup, see [our install guide for Skill Creator](/blog/how-to-install-skill-creator-claude-code-2026).

Step 3: The seven skills, in order of how often I use them

3.1 XLSX (Spreadsheet) — the one every analyst needs

This is the skill I install first on any analyst-facing machine. It lets Claude read, write, and modify .xlsx and .csv files without you opening Excel. Drop a messy CSV in your working directory, and Claude can clean it, pivot it, add formulas, and save the result, all from one prompt.

Security note: XLSX is read/write to your filesystem within Claude's working directory. Don't run it on a directory that contains files you can't afford to overwrite without backup.

3.2 ClickHouse Analytics, sub-second queries on huge tables

I tested it on a 480M-row event table. The prompt "what's the 7-day rolling DAU for the EU region, broken down by signup source" returned a working query and the chart in under 30 seconds. Same query in our existing BI tool: 6 minutes.

3.3 Supabase MCP. Postgres without leaving Claude

The 2026 release added safe-mode read-only by default. You have to explicitly opt-in to mutations, which prevents the "I asked Claude to count rows and it dropped the table" disaster that happened with earlier MCP servers.

Security note: Service-role keys give Claude admin access. Use anon keys with RLS enforced unless you specifically need privileged queries.

3.4 Neon Postgres, branching for analyst sandboxes

Neon's selling point is database branching: you can fork your production database to a free branch, do destructive analysis on it, and throw it away when you're done. The Neon Postgres skill makes this work natively in Claude.

Use case: I needed to test a denormalization strategy on a production-shaped database without touching prod. With Neon, Claude branched the DB, ran the migration, ran benchmark queries, and I had the comparison report in 12 minutes. In a non-Neon stack, this would've been a 2-day setup.

Security note: Neon API keys give full account access. Use scoped keys when available, and rotate after testing if you used a personal token.

3.5 D3.js Visualization, when Tableau isn't enough

For 80% of analyst dashboards, Tableau or Power BI is fine. The other 20% — where the chart needs to do something custom (network diagrams, force-directed layouts, interactive timelines) — D3 is still the answer. The skill knows D3 v7 syntax and can output standalone HTML files that open in any browser.

I used it last month for a customer journey visualization that needed a Sankey diagram with click-to-drill-down. Tableau couldn't do it. D3.js skill output a working prototype in two prompts.

Security note: D3 outputs are pure HTML/JS. Audit any third-party D3 templates you use — XSS via injected data is the most common failure mode.

3.6 PPTX (PowerPoint), exec decks from raw data

Half the analyst job is explaining numbers to people who don't read SQL. PPTX skill takes raw query output and produces a PowerPoint deck with charts, talking points, and an exec-summary slide. It's not pretty enough to replace a real designer, but for internal weekly reviews, it cuts deck-building from 2 hours to 15 minutes.

Security note: Filesystem write access only. Low risk.

3.7 Firecrawl, when the data lives on a website

Sometimes the dataset you need is locked in a public web page. Firecrawl scrapes it and returns clean structured data, text, tables, images, into your Claude context. It handles JavaScript-rendered pages, PDFs, and login-gated content (with cookies you supply).

I use it for competitive pricing research: scrape 30 competitor product pages, normalize the prices into a CSV, drop into XLSX skill for analysis. Whole pipeline takes ~10 minutes.

You'll need a Firecrawl API key. Free tier covers 500 pages/month.

Security note: Respect robots.txt. Most countries treat scraping public pages as legal, but TOS violations can still get you sued. Don't scrape sites you've been explicitly blocked from.

Step 4: Skills I considered but didn't include

• Notion MCP, useful for analysts who write reports in Notion, but most data work happens in spreadsheets/SQL first. Add it later if you need the Notion bridge.
• Stripe Agent Toolkit, only relevant if you're analyzing payment data via Stripe directly. Most analysts pull from Stripe's data warehouse instead.
• Frontend Design / Canvas Design, not analyst tools. Cover them in design-skill articles instead.
• Marketing Skills, overlaps too much with marketer-specific tools to fit a clean analyst stack.

Step 5: Suggested install order

For a junior or mid-level analyst starting from zero:

1. XLSX (Spreadsheet), covers 60% of daily work 2. Supabase MCP or Neon Postgres (whichever matches your DB) 3. PPTX (PowerPoint), exec decks 4. Firecrawl, only when you have a scraping task 5. ClickHouse Analytics, only if you work with high-volume event data 6. D3.js Visualization, only when standard BI isn't enough 7. Scientific Skills, for stats-heavy work (A/B test analysis, regressions)

For senior analysts and analytics engineers, swap the order: install Supabase/Neon and ClickHouse first, then XLSX. The senior workflow is database-first; spreadsheets show up at the end.

Step 6: Audit before you install

Every skill is code that runs with full filesystem access. Read the SKILL.md before installing. Specifically check:

• Does it call external APIs? Which ones?
• Does it write files outside the working directory?
• Does it require API keys, and where does it store them?
• Is the GitHub repo maintained? When was the last commit?
• Are there open security issues?

For a deeper walkthrough, see [how to audit a Claude skill before installing](/blog/how-to-audit-a-claude-skill-before-installing-2026).

Step 7: Combine skills for compound workflows

The real power isn't one skill at a time, it's chains. Example daily standup workflow I run:

1. Firecrawl scrapes overnight competitor price changes 2. XLSX combines the scraped data with our internal pricing CSV 3. Supabase MCP queries our last 7 days of order data 4. D3.js Visualization generates a comparison chart 5. PPTX wraps the chart in a 3-slide morning deck

End-to-end runtime: ~4 minutes. Manual equivalent: ~90 minutes. The skills don't have to know about each other. Claude orchestrates the chain based on the prompt.

For a similar profession-targeted list, see [our full list for data analysts](/best-ai-tools-for/data-analysts).

FAQ

Q: Are Claude skills free? The skills themselves are free (MIT or Apache licensed). You pay for Claude API usage and any third-party services the skill calls (Supabase, Neon, Firecrawl, etc.).

Q: Can I use these skills without Claude Code? Most are Claude Code-specific. Some (like the document skills) work in Claude.ai or via the API. Check each skill's SKILL.md.

Q: What's the difference between a skill and an MCP server? Skills are markdown instructions Claude reads. MCP servers are external processes Claude calls. Both extend Claude. See our [comparison](/blog/mcp-servers-vs-agent-skills-difference-2026) for the full breakdown.