pine-script-mcp

# pine-script-mcp An MCP (Model Context Protocol) server that gives any LLM full access to TradingView Pine Script v5/v6 documentation — searchable, structured, and fast. Works with **Claude Code, Claude Desktop, Cursor, Windsurf, GPT-4, Gemini**, or any MCP-compatible client. --- ## What it does Indexes the complete TradingView Pine Script documentation (User Manual + Reference) and exposes it through 5 MCP tools. Instead of an LLM hallucinating Pine Script syntax, it can query this server for accurate function signatures, examples, and conceptual docs. ## Tools | Tool | Description | |------|-------------| | `search_pine_docs` | BM25 full-text search across all docs | | `search_pine_examples` | Search and return only Pine Script **code examples** — use when writing code | | `get_pine_function` | Look up any function/variable by name (e.g. `ta.sma`, `strategy.entry`) | | `get_pine_page` | Retrieve a complete docs section (e.g. `concepts/strategies`) | | `list_pine_sections` | Browse all indexed content | | `pine_quick_ref` | Instant cheat-sheet cards — 23 topics: `indicators`, `strategy`, `plotting`, `arrays`, `inputs`, `colors`, `operators`, `types`, `time`, `loops`, `functions`, `request`, `alerts`, `drawing`, `limitations`, `matrices`, `maps`, `var_varip`, `na`, `type_qualifiers`, `libraries`, `objects`, `debugging` | ## Quick Start ```bash # 1. Clone git clone https://github.com/jrenothmisquith/pine-script-mcp cd pine-script-mcp # 2. Install dependencies pip install -r requirements.txt # 3. Install Playwright for full reference scraping (816 function docs) python -m playwright install chromium # 4. Build the docs index (one-time, ~3 minutes including Playwright) python indexer.py # 5. Start the server python server.py ``` ## Integrate with Claude Code **Option A — Project-level** (add to `.mcp.json` in your project root): ```json { "mcpServers": { "pine-docs": { "command": "python", "args": ["/absolute/path/to/pine-script-mcp/server.py"] } } } ``` **Option B — Global** (available in all Claude Code sessions): ```bash claude mcp add pine-docs python /absolute/path/to/pine-script-mcp/server.py ``` ## Integrate with Claude Desktop Edit the config file for your OS: - **Windows:** `%APPDATA%\Claude\claude_desktop_config.json` - **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json` - **Linux:** `~/.config/Claude/claude_desktop_config.json` ```json { "mcpServers": { "pine-docs": { "command": "python", "args": ["/absolute/path/to/pine-script-mcp/server.py"] } } } ``` ## Integrate with Cursor / Windsurf Add to your IDE's MCP settings (usually `mcp.json`): ```json { "servers": { "pine-docs": { "command": "python", "args": ["/absolute/path/to/pine-script-mcp/server.py"], "transport": "stdio" } } } ``` ## HTTP Mode (for remote LLMs or custom integrations) ```bash python server.py --http --port 8080 # Server available at http://localhost:8080 ``` Point any OpenAI-compatible MCP client at `http://localhost:8080`. ## What's indexed | Source | Count | Content | |--------|-------|---------| | Pine Script User Manual | ~34 pages | Language, concepts, strategies, visuals, FAQ | | Playwright SPA reference | 816 entries | Full function/variable/operator/type/constant docs scraped from the live reference SPA | | Synthetic reference | ~8 entries | Gap-fill entries for functions not caught by the SPA scraper | | Quick reference cards | 17 topics | Hardcoded cheat sheets: indicators, strategy, plotting, arrays, inputs, colors, operators, types, time, loops, functions, request, alerts, drawing, limitations, matrices, maps | **Total: ~858 documents** ## Refresh the index ```bash python indexer.py --force # re-scrape everything fresh ``` The TradingView docs update occasionally. Re-run the indexer monthly or after Pine Script version updates. ## Requirements - Python 3.10+ - Internet access for initial indexing (not required at runtime) - ~5MB disk for the index cache ## Coverage gaps & known limitations - The Playwright scraper captures what is rendered in the DOM on page load + scroll. A small number of entries may have partial descriptions if TradingView lazy-loads them beyond the scroll pass. - Pine Script v6 features may not be fully indexed if TradingView hasn't updated the reference page. - `request.security` nuances (lookahead, barmerge modes) are documented in both the scraped reference and the user manual pages. ## Optional: Playwright for full reference scraping The indexer automatically runs the Playwright scraper if `playwright` is installed. To install: ```bash pip install playwright python -m playwright install chromium python indexer.py --force ``` Without Playwright, the index still works but has ~42 docs (user manual only) instead of 858. ## Contributing PRs welcome. The most impactful improvements: 1. Fix gap-filling in `scraper_playwright.py` for entries that miss descriptions 2. Add v6 migration guide content once TradingView publishes it 3. Add more `pine_quick_ref` topics: `pine_quick_ref("pine6_changes")` ## License MIT