Most “best MCP server” lists rank whoever pays for the placement. This one states each server’s actual free-tier limit, whether it needs an API key, and what it is genuinely bad at — including ours. Limits verified 17 August 2026 against each provider’s own documentation.
These servers are not competing for the same job, and the honest answer depends entirely on your question.
Choosing the wrong category is the most common mistake. A server with excellent real-time quotes is no help if your question is “how did this behave across 2021–2026”.
| Server | Covers | API key | Free tier | Best for |
|---|---|---|---|---|
| Alpha Vantage | Stocks, options, forex, crypto, commodities, economic | Required | 25 requests/day | Breadth of asset classes |
| Polygon | US equities, options, forex, crypto | Required | Limited | Low-latency and tick data |
| Finnhub | Stocks, fundamentals, earnings, news sentiment | Required | 60 calls/min | Fundamentals and news |
| TraderMade | Forex, metals, CFDs, crypto | Required | Not published | FX specialists wanting tick data |
| MarketXLS | Equities, options chains with Greeks | Required | Paid | Options analysis |
| QuantConnect | Backtesting engine, deployment, brokerage | Required | Free tier | Running algorithmic strategies |
| FXAbsolute | 15 forex, index and crypto instruments | None | Unlimited | Deep free intraday history |
A daily request cap sounds like a minor inconvenience until you do the arithmetic for a research question.
Suppose you want five years of one-minute EURUSD bars. That is roughly 1.9 million bars. Providers page intraday history, typically returning a month or so per call. Five years is therefore around 60 calls for one instrument on one timeframe.
On a 25-requests-per-day allowance, a single instrument's history takes three days to download. Test four pairs and you have lost a fortnight before analysing anything. The cap is not really a rate limit; for historical research it is a wall.
This is the one dimension where a static archive wins outright. Files served over a CDN have no per-request accounting, so a five-year pull costs the same as a five-day one.
The part most comparisons omit.
The 25-request daily free allowance is severe for historical work. Excellent breadth otherwise — if you need equities, options, FX and economic data behind one interface, nothing else on this list matches it.
Heavily US-equities oriented. Strong for tick-level and streaming; less compelling if forex is your primary market.
The 60-calls-per-minute free allowance is the most generous here, but the strength is fundamentals, earnings and news rather than deep intraday bars.
A genuine FX specialist with tick data and built-in indicators — the closest competitor to FXAbsolute on asset focus. It requires an API key, and free-tier limits are not published, which makes planning harder.
Not a data server at heart. It exposes a backtesting engine, which is a different tool for a different job. Powerful if you write algorithmic strategies; overkill if you want to ask a question about historical ranges.
What it does better than anything else on this list is deep intraday history with nothing in the way: 28,412,683 one-minute bars spanning 2021 to 2026, no key, no cap. And it is the only one that can drive a live chart in a trader's browser, so an agent can annotate what a human is looking at rather than just returning numbers.
Nothing stops you registering more than one. They are tools, not teams, and an agent can hold several at once.
"MCP server for trading" is a claim anyone can make in a README. Three checks separate a published, maintained server from a screenshot:
https://registry.modelcontextprotocol.io/v0/servers?search=forex. A server that is not listed can still be good, but it has not been through the ownership check the registry performs.For the record, FXAbsolute's own entry: registry name io.github.varsansri/fxabsolute, npm package fxabsolute-mcp, 30 automated checks against the live archive, and an explicit statement in every relevant response that no volume field exists. fxa_session_scan additionally reports how many weekdays in a requested window produced no data at all — SPX500 is missing about 19% of weekdays in the first half of 2024, and a scan that quietly omitted that would return clean-looking statistics built on a fifth less data than you asked for.
Query the official registry at https://registry.modelcontextprotocol.io/v0/servers?search=<term>. Listed servers have passed an ownership check tying the entry to the publisher. Also check whether it installs from a package registry like npm rather than a raw GitHub branch.
It depends on the question you are asking. Alpha Vantage has the broadest asset coverage; Polygon is strongest for low-latency US equity and tick data; Finnhub is best for fundamentals and news; FXAbsolute offers the deepest free intraday history with no API key. There is no single winner across all four needs.
FXAbsolute requires no API key and applies no rate limit, because its candle data is served as static files. Alpha Vantage, Polygon, Finnhub, TraderMade, MarketXLS and QuantConnect all require registration and a key.
As documented on 17 August 2026, the Alpha Vantage free tier allows 25 API requests per day, with unlimited access available for verified open-source or educational projects. That cap is restrictive for downloading long intraday histories.
Some can. Polygon and TraderMade offer real-time and streaming data. FXAbsolute does not — it is a historical archive covering 2021 to 2026, intended for research and backtesting rather than live execution.
Genuine forex volume does not exist centrally, since FX trades over the counter rather than on a single exchange. Providers that report forex volume are reporting their own feed’s tick counts. FXAbsolute omits the field entirely rather than publishing a number that could be mistaken for market-wide volume.
FXAbsolute is a free browser backtester running on the same archive the MCP server exposes. No account, no download.
Open FXAbsolute →