TickDB: one market data API for six markets, with the MCP server open sourced

Six markets behind one symbol grammar

The problem TickDB addresses is vendor sprawl in market data. Instead of one provider per asset class, a single API covers forex pairs such as GBPUSD, precious metals such as XAUUSD, US stocks written AAPL.US for NYSE and NASDAQ listings, indices such as SPX, HK stocks such as 700.HK, A-shares such as 600519.SH for the Shanghai and Shenzhen exchanges, and crypto pairs such as BTCUSDT. The README claims 37,527-plus symbols across six markets, a vendor figure that should be checked against the /v1/symbols/available endpoint rather than taken on faith. The data shapes are the standard trio: candlesticks, tick-level trades, and order book depth. Who is this for? The README names quantitative trading, dashboards, trading applications, backtesting, and AI agents, and the honest core of that list is the cross-market cases, since a US-only or crypto-only project has well-established alternatives with deeper per-market coverage. One connection that spans A-shares and NASDAQ in one schema is the actual product.

What the repository open sources, and what it does not

Reading the repository needs one clarification up front: the code here is client tooling and documentation, while the market data service is TickDB's hosted, commercial product at api.tickdb.ai, with a free key available without a credit card. The repository provides SKILL/, the AI skill configuration; mcp/, a Python MCP server described as 13 tools with a Dockerfile, 46 unit tests, CI, and an MIT licence; and the documentation that drives the docs-quality workflow badge. The README makes a transparency claim worth quoting in substance: the hosted endpoint mcp.tickdb.ai runs the same code as mcp/, so what you read, what you can deploy, and what the vendor operates are stated to be identical. That claim is checkable by deploying the container yourself, and it is the right kind of promise for a data vendor to make about its tooling. It does not change the fact that the data, pricing tiers beyond free, and uptime all belong to the vendor rather than the repository.

REST calls you can copy today

The REST surface is small and concrete. Authentication is a single header, X-API-Key, on every request to the base URL https://api.tickdb.ai. Five endpoint patterns carry the load: /v1/market/kline for candlesticks with symbol, interval and limit parameters, /v1/market/ticker for quotes across comma-separated symbols, /v1/market/depth for order book levels, /v1/market/trades for recent executions, and /v1/symbols/available for discovery filtered by market. The README's examples are directly runnable once a key exists, for instance curl -H "X-API-Key: YOUR_API_KEY" "https://api.tickdb.ai/v1/market/kline?symbol=700.HK&interval=1h&limit=24", and a ticker query across AAPL.US,700.HK,BTCUSDT demonstrates the multi-market pitch in one call. Nothing in the material documents rate limits, pagination behaviour, or historical depth per endpoint, so those belong on the verification list before integration.

WebSocket streaming, and an API key sitting in the URL

Real-time delivery is WebSocket-based, at wss://api.tickdb.ai/v1/realtime, with three channels: ticker for price updates, depth for order book changes, and trade for executions. The subscription protocol is a JSON message with a subscribe command and a data object naming the channel and symbols. The README's latency claim for this stream is 10 to 50 milliseconds end to end, which is a vendor figure with no methodology attached, so treat it as a target to reproduce rather than a specification. One design detail deserves attention: the example URL passes the API key as a query parameter, api_key=YOUR_API_KEY, presumably because browser WebSocket clients cannot set custom headers. Keys embedded in URLs routinely end up in server logs, proxy logs and analytics pipelines. For a REST call the key sits in a header, which is cleaner; teams with strict secret-handling rules should weigh the WebSocket pattern before putting production keys on that connection.

Three AI tiers: a trial-key Skill, hosted MCP, and self-deployed MCP

The AI-native positioning is where TickDB differs most from incumbent data vendors. The first tier is the Skill: npx clawhub@latest install tickdb-market-data installs a skill that the README says automatically obtains a trial key, letting an AI chat query 72 popular symbols with no registration at all. The second tier is MCP: a one-time JSON config pointing at the hosted endpoint https://mcp.tickdb.ai/ with an X-TickDB-Key header gives Claude Code, Claude Desktop, Cursor, Kiro, Codex, Zed or Cherry Studio 13 market-data tools permanently. The third tier is self-deployment of the mcp/ directory, with mcp/MCP_CLIENT_SETUP.md and mcp/DEPLOYMENT.md documenting client configuration and hosting, aimed at teams that want the data flow under their own control. A CLI completes the set: npm install -g tickdb, then tickdb config set-key YOUR_API_KEY, then tickdb ticker BTCUSDT,XAUUSD. The tiering is coherent, from zero-config trial to self-hosted integration, and it maps cleanly onto how AI-assisted development actually consumes APIs today.

A quiet summer, vendor numbers, and the alternatives

Maintenance signals are modest but present. The last push was 2026-06-20, and the MCP server's releases are mcp/v0.1.2 on 2026-05-22 and mcp/v0.1.3 on 2026-06-20, roughly three months of quiet before this review's reference date, with CI and docs workflows still badge-displayed. The licence is MIT for the repository's tooling, while the service terms live on the vendor's site, and that split, MIT client code over a proprietary feed, is the standard commercial open-source-adjacent arrangement rather than a red flag. Alternatives differ by need. For crypto only, CCXT is a genuinely open-source library that connects directly to exchange APIs with no middleman vendor, trading TickDB's unified streaming for zero subscription dependency. For US equities depth and options, single-market specialists such as Polygon.io go deeper than a six-market aggregator reasonably can. TickDB's defensible slice is the combination: A-shares and HK stocks beside US names and crypto in one schema, reachable by an AI client through MCP in an afternoon. Whether that slice is worth a recurring vendor relationship depends on pricing the material does not show.

Editorial conclusion

Developers who need US, HK and A-share equities next to forex, metals and crypto in a single API, and who want AI clients to query it directly, should register a free key at tickdb.ai and reproduce the README's curl examples for /v1/market/ticker before committing. It is the wrong choice for anyone needing free open data, deeper single-market coverage such as full US options chains, or a self-hosted data backend, since the service itself is hosted and commercial even though the mcp/ tooling is MIT licensed. Verify three things first: pricing and limits beyond the free tier, which the material does not specify, the 10 to 50 ms latency claim from your own region, and whether the WebSocket's api-key-in-URL pattern passes your secret-handling policy before anything reaches production.

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. TickDB/tickdb-unified-realtime-marketdata-api on GitHub