DEV Community

vdalhambra
vdalhambra

Posted on

I built 2 MCP servers that make Claude a financial analyst and SEO auditor

#mcp #claudecode #python #ai

If you use Claude Code or Claude Desktop, you probably know about MCP (Model Context Protocol) — the standard for giving AI agents access to external tools. There are 17,000+ MCP servers out there, but most are basic API wrappers.

I built two that try to go beyond wrapping: they add intelligence on top of the data.

FinanceKit MCP — Financial Market Intelligence

12 tools. The core value is the technical analysis engine — it doesn't just return raw data from Yahoo Finance. It calculates 10 technical indicators, detects patterns, and gives you a plain-English verdict.

Ask Claude: "Run technical analysis on NVDA"

You get:

  • Overall bias: STRONGLY BULLISH (quantified: 3.0 bullish vs 0.5 bearish signals)
  • RSI, MACD, Bollinger Bands, SMA(20/50/200), EMA, ADX, Stochastic, ATR, OBV
  • Pattern detection: Golden Cross, Death Cross, overbought/oversold
  • 7 detailed signals like: "MACD bullish — line above signal, both positive. Strong upward momentum."

Other tools: stock quotes, crypto prices (CoinGecko), market overview (S&P 500, NASDAQ, DOW, VIX, top movers), multi-asset comparison with Sharpe ratios and drawdown, portfolio analysis with sector breakdown.

SiteAudit MCP — Instant Website Audits

8 tools. Give it any URL and it runs 20+ SEO checks, analyzes security headers, verifies SSL certificates, measures performance, and can compare multiple sites.

Ask Claude: "Audit stripe.com"

You get:

  • Overall: 92/100 (Grade A) — SEO: 94, Performance: 90, Security: 90
  • Specific issues: "Title too long (66 chars)", "46/75 images missing alt text"
  • SSL certificate status and expiration
  • Performance: 66ms response, gzip compression, no redirects

The compare_sites tool is the killer feature for competitive analysis — compare your site against 3 competitors in one command.

Also includes Lighthouse scores and Core Web Vitals via Google PageSpeed Insights API, and broken link detection with concurrent checking.

How to install

⭐ Recommended: MCPize (zero setup, free tier)

The fastest path. No terminal, no Python setup, no config wrestling. Works immediately in Claude Desktop, Cursor, Windsurf, and Claude Code:

👉 Install FinanceKit on MCPize — Free 100 calls/mo
👉 Install SiteAudit on MCPize — Free 100 audits/mo

Or add directly to your MCP config:

{
  "mcpServers": {
    "financekit": { "url": "https://financekit-mcp.mcpize.run/mcp" },
    "siteaudit": { "url": "https://siteaudit-mcp.mcpize.run/mcp" }
  }
}
Enter fullscreen mode Exit fullscreen mode

Pricing tiers

FinanceKit: Free ($0) → Hobby ($9) → Pro ($29) → Team ($79) → Business ($179) → Enterprise ($499)

SiteAudit: Free ($0) → Hobby ($7) → Pro ($19) → Agency ($49) → Agency Plus ($119) → Enterprise ($349)

Bundle: Pro Combo (both) = $39/mo (save 19%)

Annual plans get 2 months free.

Advanced: self-hosted (developers)

Prefer to run locally? Both are MIT-licensed and free to self-host — you manage updates and rate limits:

Self-hosted install

# Claude Code CLI
claude mcp add financekit -- uvx --from financekit-mcp financekit
claude mcp add siteaudit -- uvx --from siteaudit-mcp siteaudit

# Or Claude Desktop / Cursor / Windsurf config
{
  "mcpServers": {
    "financekit": {
      "command": "uvx",
      "args": ["--from", "financekit-mcp", "financekit"]
    },
    "siteaudit": {
      "command": "uvx",
      "args": ["--from", "siteaudit-mcp", "siteaudit"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Also on Smithery (one-click install) and PyPI.

For most users, MCPize is the better choice — zero setup, always up-to-date, no infrastructure to manage.

Tech stack

  • FastMCP 3.2 for the MCP server framework
  • yfinance + CoinGecko API for financial data
  • ta library for technical analysis calculations
  • BeautifulSoup for HTML parsing (SiteAudit)
  • Google PageSpeed Insights API for Lighthouse (free, 25K calls/day)
  • TTL-based caching to minimize API calls

Both are open source (MIT) and now listed on the Official MCP Registry (io.github.vdalhambra/financekit-mcp and siteaudit-mcp):

v1.2 is live — 8 new premium tools

Just shipped v1.2.0 with 8 new premium tools (Pro tier+):

FinanceKit (+5 tools, now 17 total):

  • risk_metrics — VaR (95%), Sharpe, Sortino, Beta vs benchmark, Max Drawdown
  • correlation_matrix — Pairwise correlations + diversification score
  • earnings_calendar — Next earnings + EPS estimates vs reported
  • options_chain — Calls/puts with volume, open interest, implied volatility
  • sector_rotation — 11 GICS sectors ranked by performance

SiteAudit (+3 tools, now 11 total):

  • accessibility_audit — WCAG checks: alt text, form labels, heading hierarchy, ARIA, landmarks
  • schema_validator — Extract + validate Schema.org JSON-LD with required-field checks per type
  • competitor_gap_analysis — Audit your site vs up to 5 competitors, returns gaps + priority focus

More coming in v1.3: smart alerts, backtesting, full-site crawler, white-label PDF reports, scheduled audits.

Support the project

If these tools are useful to you:

What MCP servers are you building or using? Would love to hear about interesting use cases.

Top comments (2)

Collapse
 
automate-archit profile image
Archit Mittal •

This is exactly the kind of practical MCP use case that shows why the protocol is taking off. Turning Claude into a domain-specific analyst by giving it structured access to financial data and SEO metrics is way more powerful than just pasting data into prompts. The key advantage I see with the MCP approach vs. custom function calling is that these servers become reusable across projects - build once, use everywhere. Have you considered adding caching at the MCP server level? For financial data that doesn't change intraday, caching API responses can significantly reduce latency and costs.

Collapse
 
vdalhambra profile image
vdalhambra •

Exactly on the reusability front — that's the MCP advantage that's hard to convey until you try it. Same server config works in Claude Desktop, Cursor, Windsurf, and Claude Code without any changes.

On caching: already there at the server level with TTL-based caching (listed in the tech stack). Stock quotes cache for a few minutes, technical analysis results longer since they're computed from OHLCV data that doesn't change mid-candle. For portfolio queries across 10 tickers, individual ticker responses are cached so you're not hitting yfinance 10x per query.

Next improvement: smarter TTL tied to market hours — shorter during live sessions, longer overnight. Simple but high-impact for anyone running the server persistently.