tradingview-mcp

Screeners, technical analysis and backtests for any MCP client. No TradingView account.

by Ahmet Taner Atila

Last updated

From
$9/mo
Free tier
Yes
Latency
Delayed
Non-US markets
Yes

What it is

A Python MCP server that hands an AI client 37 market tools: screeners across Binance, KuCoin, NASDAQ, NYSE, BIST and the Egyptian Exchange, technical analysis (RSI, MACD, Bollinger and 23 indicators with a BUY/SELL/HOLD summary), US options chains with a volume-over-open-interest unusual-activity scan, licensed news and per-symbol sentiment, and a backtest engine running nine fixed strategies with Sharpe, Calmar, profit factor and a walk-forward overfitting verdict.

It is not a TradingView product and gets none of TradingView's licensed data. The author states that the project is not affiliated with, endorsed by or associated with TradingView Inc.; it needs no TradingView account, key or plan, and it does not drive a logged-in TradingView session the way the desktop-automation servers of the same name do. The name describes one of the public endpoints it reads, not who wrote it.

Pricing

Self-hosting is free under MIT. The same author sells a hosted connector branded CryptoSieve — one URL, no ops — at $9 a month for 2,500 calls and 60 requests a minute, or $29 for 10,000 calls and 150 a minute; yearly Pro is $90. The trial is three days and takes a card. The hosted product advertises 27 tools where the repo lists 37.

Data & coverage

No data of its own and no exchange licence. Quotes come from Yahoo Finance's public endpoints, as do the options chains and the unusual-activity ranking; screening and indicators come from TradingView's public screener via the tradingview-screener and tradingview_ta packages. News and per-symbol sentiment come from Marketaux, a licensed news API — one fetch serves both tools, the server caches it for four hours and caps itself at 90 calls a day, and without a Marketaux token those two tools return a "not configured" stub rather than data. Each source carries its own terms and its own reliability: crypto is effectively live, equity venues lag, and an upstream change can silence a tool — news and sentiment have been rebuilt twice in 2026, first off Reuters' deprecated RSS feeds and then off the Reddit and RSS scrapers entirely.

Limitations

Read-only analysis: no order routing, no paper trading, no portfolio, no alerts. Backtests are nine presets on daily or hourly bars, not your own strategy code. The server caps itself at four concurrent analysis calls with 0.8-second spacing because TradingView's endpoints rate-limit it, so wide scans are slow. Publishing is irregular: the last tag is v0.8.0 (29 July 2026), PyPI's latest is 0.8.1 (24 August 2026), the README already badges v0.9.0, and the last commit to main was 1 September 2026. Check the package name before you install: this project publishes as tradingview-mcp-server, while the bare name tradingview-mcp on PyPI belongs to an unrelated single-release package by another author — and it is numbered 0.9.1, higher than the real one, so the version number will not warn you. Several tools emit verdicts on a named ticker rather than data: multi_agent_analysis returns STRONG BUY through STRONG SELL with a confidence label, and the EGX screener and trade-plan tools attach entry points, a stop loss, targets and a trade-quality score. It is indicator arithmetic, and the README disclaims it as not advice, but it is a call and not a number — which is why this card carries the recommendations flag.

Alternatives

Vendor-official MCP servers from Alpha Vantage, EODHD or Massive if you want licensed data and someone to complain to. OpenBB for a research workbench rather than a screener bus. If what you wanted was an AI clicking around your own TradingView Desktop — Pine Script, drawings, bar replay — that is the unrelated tradesdontlie/tradingview-mcp project, which does need your TradingView account.

Specs

Interfaces
Python, MCP server
Export
JSON
Asset classes
Stocks, ETF, Options, Crypto, Forex, Futures, Indices
Markets
Global
Platforms
CLI, Library
AI features
Recommendations
Capabilities
Screening, Scanning, Backtesting, News, Options analysis
Pricing verified
Capabilities verified
Coverage verified

Also worth comparing

  • EquiblesScrapes SEC, FINRA, FRED and CFTC data into your own Postgres and serves it over MCP.
  • Financial DatasetsUS fundamentals and filings for agents, over REST and a first-party hosted MCP server.
  • OpenBBOne Python API, REST app and MCP server over thirty other people's data feeds.
  • MaverickMCPThirty-seven stock-analysis tools for any MCP client, on Yahoo data and no key.
  • TradingAgentsAnalyst, bull, bear, risk and portfolio-manager agents argue over one ticker.
  • Vibe-TradingA local finance research agent with 74 MCP tools, nine backtest engines and 14 brokers.

FAQ

Is this made by TradingView?

No. It is an independent open-source project by Ahmet Taner Atila, and the README says so in as many words — not affiliated with, endorsed by or associated with TradingView Inc. TradingView sells its own charting platform and has no part in this server.

Do I need a TradingView account or a paid TradingView plan?

No. The server needs no TradingView account, login or API key, and it does not automate a browser session. That distinguishes it from the desktop-automation servers sharing the name, which remote-control TradingView Desktop and do need your logged-in account.

Where does the data actually come from?

Yahoo Finance public endpoints for quotes and for US options chains, and TradingView's public screener through the open-source tradingview-screener and tradingview_ta packages for screening and indicators. News and per-symbol sentiment come from Marketaux, a licensed news API, and the two tools that use it return a "not configured" stub until you set a free Marketaux key — 100 requests a day. The Reddit scraper and the RSS reader earlier versions used are still in the package but nothing imports them any more. None of it is licensed exchange data.

Is it free?

Self-hosting is free under MIT and always will be. The same author sells a hosted connector as CryptoSieve at $9 or $29 a month for people who would rather not run it, with a three-day trial that requires a card.