TradingView Indicator Search MCP Server

Search TradingView’s public indicator and strategy library from Claude, Cursor, or any MCP client, and pull the full Pine Script source of any open-source script.

An MCP (Model Context Protocol) server that turns the entire public TradingView script library into a tool your AI coding agent can query. Ask for “trend following strategies” or “everything by LuxAlgo”, get back real indicators with their Pine Script source, ready to read, adapt or backtest.

No TradingView account, API key, cookie or login required. Every endpoint used is public.

---

Contents

• What this does
• Quick start
• Install in Claude Code
• Install in Claude Desktop
• Install in Cursor
• Tools
• Example prompts
• How TradingView search actually works
• Rate limiting and not getting banned
• Configuration
• FAQ
• Development

---

What this does

TradingView hosts hundreds of thousands of community-published indicators and strategies written in Pine Script. Most are open source. This MCP server gives an AI agent structured access to all of it:

• Search by topic — “trend following”, “scalping”, “order blocks”, “mean reversion”
• Browse editors’ picks — TradingView’s hand-curated best scripts
• Search by author — every public script from a creator you follow
• Fetch Pine Script source — the complete code for any open-source script
• Build a local library — bulk-collect scripts into a SQLite corpus you can search offline, including searching inside the Pine code itself

That last point is the useful one. Once scripts are collected locally, you can find every strategy that uses ta.supertrend or strategy.entry with a single instant query. The TradingView website cannot do that.

Quick start

git clone https://github.com/daviddme/tradingview-indicator-search-mcp-server.git
cd tradingview-indicator-search-mcp-server
npm install

Requires Node.js 22.5 or newer (it uses the built-in node:sqlite, so there is no native module to compile and no database to install).

Then add it to your MCP client below.

Install in Claude Code

claude mcp add tradingview -s user -- node "$(pwd)/src/server.js"

-s user makes it available in every project. Verify with:

claude mcp list

Install in Claude Desktop

Edit claude_desktop_config.json:

• macOS — ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows — %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "tradingview": {
      "command": "/absolute/path/to/node",
      "args": ["/absolute/path/to/tradingview-indicator-search-mcp-server/src/server.js"]
    }
  }
}

Use the absolute path to node. Desktop apps do not inherit your shell PATH, so a bare "node" fails with spawn node ENOENT. Find yours with which node.

Restart Claude Desktop.

Install in Cursor

Edit ~/.cursor/mcp.json with the same block as above, then restart Cursor.

Any MCP-compatible client works. The server speaks stdio, so it also runs in Windsurf, Cline, Zed and others.

Tools

Example prompts

Once installed, just ask your agent in plain language:

“Find me the top 10 trend-following strategies on TradingView and show me the Pine Script for the most popular one.”

“Harvest 100 scalping strategies, then show me which ones use ATR-based stop losses.”

“What indicators has LuxAlgo published recently?”

“Get the source for https://www.tradingview.com/script/K0SEi3Ct-Some-Indicator/ and explain how its entry logic works.”

“Search my local corpus for every strategy that calls strategy.exit with a trailing stop.”

The efficient pattern is to harvest once, then work offline:

harvest_scripts   query="trend following" script_type=strategies max_scripts=100
query_corpus      source_contains="strategy.entry" with_source_only=true
get_script_source script_id_part="PUB;..."

How TradingView search actually works

Worth knowing, because it explains the tool design. TradingView exposes two search surfaces and they are not equally good:

So topic search uses the slug routes, and the JSON API is used only for by= author lookups, which is the one thing it does well.

Undocumented filters compose with topics, and the API helpfully declares its own query schema in every response under data.ideas.params:

• script_type — indicators | strategies | libraries | all
• script_access — open | all
• sort — latest_popular | recent_extended

Every script carries a script_id_part (PUB;) and version, which together address the Pine source endpoint. Passing last as the version always resolves to the current one.

Rate limiting and not getting banned

This is the part most scrapers get wrong, so it is enforced at the lowest level: every outbound request passes through a single throttle and no tool can bypass it.

• Strictly sequential — concurrency of exactly 1, process-wide. Never parallel
• Randomised delays of 1.5 to 4.5s between requests. A fixed interval is the clearest bot signature there is, so the gap is jittered every time
• Human break pattern — a 20 to 60s pause roughly every 25 requests
• Persistent cookie jar — real browsers accumulate cookies, and sending none on every request is itself anomalous. The server warms up once per 12h and reuses the jar
• Stable browser identity — user agent and its matching sec-ch-ua are chosen once per process and held constant, because real browsers do not change UA mid-session
• Exponential backoff on 429/403/5xx, honouring Retry-After
• Circuit breaker — after 3 consecutive blocks, all traffic stops for 15 minutes and tools return a clear error. Without this, a rate-limited agent retries straight into a hard ban. This is the single most important safeguard here
• Daily budget of 2000 requests, so a runaway agent loop cannot burn your IP overnight
• Cache-first — Pine source is immutable per version, so a cached source is never refetched. This cuts request volume more than everything else combined

Run fetch_status at any time to see budget use and breaker state.

Run this locally, on a residential connection. Hosting it on a cloud provider puts requests on a datacenter IP, which sites like TradingView block far more aggressively than home connections, and a block would hit an egress IP you do not control.

Configuration

All optional, all environment variables.

Running several MCP clients at once is fine. They share one SQLite file (WAL mode, so concurrent access is safe) and therefore share the daily request budget, which is correct because they share one IP.

FAQ

Do I need a TradingView account or API key?

No. Every endpoint this uses is public and unauthenticated. There is no login, no API key and no cookie to configure.

Can it fetch invite-only or closed-source indicators?

No, and nothing can. TradingView never exposes the Pine source of protected scripts. Searches filter them out by default so every result you see can actually be fetched.

Does this work with Claude Desktop, Claude Code and Cursor?

Yes, all three, plus any other MCP client that speaks stdio. See the install sections above. It does not work with claude.ai in the browser, which requires a publicly hosted HTTPS connector rather than a local process.

Will this get my IP banned by TradingView?

The whole design is built to avoid that: sequential requests, randomised human-like delays, periodic breaks, a daily budget and a circuit breaker that halts everything after repeated blocks. Run it locally rather than from a datacenter, and leave the conservative defaults alone unless you have a reason.

How many scripts can I collect?

TradingView caps broad result counts at 1000, so narrow topics give better coverage than broad ones. A single harvest_scripts call is capped at 200 scripts by default, and it is resumable, so larger libraries are built across several calls.

Where is the data stored?

In a local SQLite file, data/indicator_fetch.db by default. Nothing is uploaded anywhere. Change the location with IF_DB_PATH.

Can I search inside the Pine Script code?

Yes, and this is the main reason to build a local corpus. query_corpus supports full-text search over names, authors and descriptions, plus literal substring matching inside the source with source_contains, which is what you want for code tokens like ta.supertrend that a full-text tokeniser would split apart.

Is scraping TradingView allowed?

This reads public pages the same way a browser does, at a much slower rate than a person clicking around. You are responsible for your own use of it, including TradingView’s Terms of Service, and for respecting the licences of any Pine Script you collect. Script authors retain copyright in their code.

Development

npm test

60 tests, fully offline. Parsers run against real captured API payloads in test/fixtures/ (see the fixtures note on what was redacted for redistribution), and the throttle is tested with an injected clock so no test ever waits.

npm run smoke

Live end-to-end check against TradingView. Makes about 5 real requests.

Project layout

License

MIT. See LICENSE.

Pine Script collected with this tool remains the copyright of its original authors on TradingView. This project is not affiliated with or endorsed by TradingView. See NOTICE.md.

---

Keywords: TradingView MCP server, TradingView MCP, Model Context Protocol TradingView, Pine Script MCP server, TradingView indicators API, Pine Script scraper, TradingView strategy search, MCP server for trading, Claude TradingView integration, Cursor TradingView MCP, AI trading indicators, TradingView Pine Script source code, backtesting indicators with AI.