Case Study: The DailyTickers MCP Server

What is DailyTickers?

DailyTickers is a financial analysis platform that publishes institutional-quality market reports — daily briefings, weekly outlooks, individual stock analyses, and algorithmic scanner picks. Every article is generated by an AI agent powered by Claude Code and a custom MCP server.

The Problem We Solved

Financial analysis requires pulling data from dozens of sources — stock prices, options chains, earnings calendars, economic indicators, news, sentiment. Before MCP, each analysis required manually gathering this data. With our MCP server, Claude Code can autonomously:

7 Tools + 1 Resource

The DailyTickers MCP Gateway exposes 7 carefully designed tools and 1 resource. Each tool is optimized for a specific workflow in the article generation pipeline.

Resource: marketwatch://watchlist

A read-only resource exposing the current watchlist — a curated list of symbols the platform actively tracks. Claude reads this resource to know which stocks to include in daily briefings and weekly reports.

QueryData — The Swiss Army Knife

QueryData is the most-used tool because it supports 58 different data types through a single interface. The AI specifies which data types it needs, and the server fetches all of them in parallel.

// Claude calls QueryData to get everything about NVDA
{
  "types": "quote,bars_daily,financials,earnings_quarterly,holders,stats,support_resistance,volume_profile,sentiment_overall,trading_signals,analyst_actions,insider_transactions,news",
  "symbols": "NVDA"
}

// Returns a unified JSON with all 13 data types
// Each type is keyed by its name:
{
  "quote": { "price": 142.50, "change": 3.2, ... },
  "bars_daily": [{ "date": "2026-03-05", "close": 142.50, ... }],
  "financials": { "revenue": 130.5e9, ... },
  "sentiment_overall": { "score": 0.72, "label": "bullish" },
  ...
}

Data Flow — From API to Published Article

The Pipeline

Typical Tool Call Sequence for a Daily Briefing

// 1. Market snapshot
GetMarketOverview({ depth: "deep" })

// 2. US indices + major ETFs
QueryData({
  types: "quote,bars_daily",
  symbols: "SPY,QQQ,DIA,IWM,GLD,SLV,USO,TLT"
})

// 3. International markets
QueryData({
  types: "quote",
  symbols: "EFA,EEM,FXI"
})

// 4. Crypto
QueryData({
  types: "quote",
  symbols: "BTC-USD,ETH-USD,SOL-USD,XRP-USD"
})

// 5. Web search for context
WebSearch("CPI inflation March 2026 FOMC")
WebSearch("earnings calendar this week")
WebSearch("geopolitics Ukraine China tariffs")

// Total: 3 MCP calls + 3 web searches = complete daily briefing

Automated Publishing Pipeline

DailyTickers runs on a fully automated schedule. A Discord bot manages task scheduling, and Claude Code does the heavy lifting.

The Discord Bot Architecture

// Schedule management via Discord
every day at 07:00 articles analyse daily
every weekday at 23:00 articles scan du jour
every friday at 23:00 articles retrospective scanner
every sunday at 18:00 articles nouvelle analyse weekly

// Manual commands
articles analyse NVDA         // Generate ticker analysis
articles scan du jour          // Run scanner now
schedules                      // List all scheduled tasks
run #1                         // Execute task #1 immediately

The bot translates Discord commands into claude -p "command" executions in a tmux session. Claude Code reads the project's CLAUDE.md (which contains all the article generation instructions), calls MCP tools for data, generates the HTML, runs the indexing script, and git pushes.

Lessons Learned — What Works, What Doesn't

What Works Brilliantly

• Structured data in, structured article out. When MCP tools return clean JSON, Claude produces consistent, high-quality articles every time.
• CLAUDE.md as the instruction manual. Detailed instructions in the project's CLAUDE.md file act as the "brain" — the AI follows them reliably for 50+ section templates.
• Reference articles. Telling Claude "read this existing article and match the format" produces better results than describing the format abstractly.
• Git as deployment. git push triggers GitHub Pages deployment — no CI/CD configuration needed.
• ECharts for every chart. Claude is remarkably good at generating ECharts configurations from raw data.

Anti-Hallucination Strategies

The biggest risk with AI-generated financial content is hallucinated data. Here's how we prevent it:

What Doesn't Work

• Asking Claude to "remember" market data. It will confidently cite wrong prices from training data. Always fetch live data via MCP.
• Too many tools at once. When we had 25+ tools, Claude would sometimes pick the wrong one. We consolidated into QueryData with types.
• Inline CSS everywhere. Articles initially had thousands of inline styles. Migration script cleaned 433 articles to use shared CSS classes.
• No validation post-generation. Early articles sometimes had broken HTML, missing sections, or wrong dates. Adding size checks and section counters fixed this.

The Future of DailyTickers MCP

Near-Term Roadmap

The Bigger Picture

DailyTickers demonstrates a pattern that will become ubiquitous: AI agents + MCP servers + automated workflows = autonomous content production. The same architecture works for any domain — legal research, medical literature reviews, competitive intelligence, technical documentation.

The key ingredients are:

• A well-designed MCP server that provides reliable, structured data
• Detailed instruction files (CLAUDE.md) that encode domain expertise
• Reference articles that demonstrate the expected output format
• Validation pipelines that catch errors before publication
• A scheduling system that runs workflows autonomously

Series Capstone Quiz