Tickerdb MCP Server

TickerDB — Stock market data for agents.

Pre-computed stock market data for AI agents. TickerDB returns indicators like trend_direction, support_level, and analyst_consensus as named states — plus what changed and what usually happens next.

10,000+ US stocks, ETFs, and crypto pairs · 182 indicators across trend, momentum, volatility, volume, patterns, support/resistance, fundamentals, and sector context · 7 years of history · tickerdb.com

Tools

All tools are available on every tier (Free, Plus, Pro). Tiers differ by credit limits, history depth, number of filters, and watchlist size. See tickerdb.com/pricing.

Quick start

Connect TickerDB to Claude, ChatGPT, or another MCP client (see Setup below), then try:

"Show me oversold large-cap stocks near support"

The agent calls get_search with filters for momentum_rsi_zone = oversold and market_cap_tier in [large, mega], then follows up with get_summary on individual results. No raw number crunching — the agent reads categorical states and reasons over them directly.

"What usually happens when AAPL goes oversold?"

get_summary with field=momentum_rsi_zone, band=oversold, stats=true returns aggregate aftermath distributions: how the stock performed 5, 10, 20, 50, and 100 days after each oversold entry over 7 years of history.

"What changed on my watchlist?"

get_watchlist_changes returns only the field-level state transitions since the last pipeline run — band entries, exits, and shifts — so the agent reports what moved without pulling full summaries for every ticker.

Why not just pass raw OHLCV?

A model can compute RSI from raw bars. But ask "Does AAPL look bullish?" with raw OHLCV and it burns its context on arithmetic — computing indicators one by one — instead of doing what you actually asked: noticing that RSI just hit oversold while institutions are accumulating, that the pullback is sharp but the 200-day uptrend is intact, that insiders have been selling all quarter. That's the analysis. Raw bars bury it under computation.

With TickerDB, the model sees "oversold", "accumulation", "strong_uptrend" and connects them immediately.

State transitions go further. "What happened the last time BTC was this oversold?" means computing RSI across 7 years of daily bars, finding every oversold entry, and calculating what happened after each one. With TickerDB it's one call: get_summary with field=momentum_rsi_zone, band=oversold, stats=true.

Setup

Hosted server (recommended)

The remote server at https://mcp.tickerdb.com/mcp supports OAuth 2.1 and Bearer token auth. Use Streamable HTTP transport (not legacy SSE).

npm package (local stdio)

For clients that prefer a local process (Claude Desktop, etc.):

{
  "mcpServers": {
    "tickerdb": {
      "command": "npx",
      "args": ["tickerdb-mcp"],
      "env": {
        "TICKERDB_KEY": "tdb_your_api_key_here"
      }
    }
  }
}

Get an API key at tickerdb.com/dashboard.

Structure

Three-package monorepo:

• shared/ — Tool definitions, API client, and server factory (internal)
• remote/ — Cloudflare Worker at mcp.tickerdb.com (Streamable HTTP + OAuth 2.1)
• local/ — Published npm package tickerdb-mcp (stdio)

Both transports use the same tool definitions. The MCP server is a thin proxy — access control, rate limiting, and field filtering are handled by the TickerDB API.

Authentication

• Bearer token — Authorization: Bearer tdb_...
• OAuth 2.1 — dynamic client registration, PKCE, token exchange, revocation. /authorize redirects to tickerdb.com for consent.

Unauthenticated initialize and tools/list are permitted for tool discovery; tools/call requires auth and returns a 401 Bearer challenge with resource_metadata for clean re-authorization.

Session strategy

The remote worker defaults to stateless transport — intentionally. All tools are request/response stateless, and Cloudflare Worker memory is isolate-local. Stateless mode avoids edge session loss that can invalidate connector-discovered namespaces. Set MCP_SESSION_MODE=stateful for explicit session debugging.

Development

npm install                              # workspace dependencies
npm run build                            # type-check remote + shared
npx wrangler dev                         # remote dev server
cd local && npm install && npm run build  # npm package

Deployment

# Remote server
npx wrangler deploy

# npm + MCP Registry (recommended)
export MCP_PUBLISHER_KEY="your_saved_tickerdb_registry_private_key_hex"
./release.sh mcp patch

# npm only
cd local && npm version patch && npm run build && npm publish