{"filename":"helium-mcp-cursorrules-prompt-file.mdc","content":"---\ndescription: \"Cursor rules for Helium Mcp.\"\nglobs: **/*\nalwaysApply: false\n---\n# Helium MCP in Cursor\n\nWhen the user’s project or question involves **news**, **media bias**, **markets**, **options**, or **financial memes**, prefer **Helium MCP tools** (if enabled) over guessing headlines, prices, or Greeks. Helium is a hosted MCP service from [Helium Trades](https://heliumtrades.com); upstream docs and source live at [github.com/connerlambden/helium-mcp](https://github.com/connerlambden/helium-mcp).\n\n## Enable the server in Cursor\n\nAdd a streamable HTTP MCP entry pointing at the hosted endpoint (no API key required for the free tier; paid tiers use keys from the MCP page):\n\n```json\n{\n  \"mcpServers\": {\n    \"helium\": {\n      \"url\": \"https://heliumtrades.com/mcp\"\n    }\n  }\n}\n```\n\nAlternative: stdio bridge via `npx @mcp-get-community/server-helium` if your workflow requires a local process (see that package’s README for `HELIUM_MCP_URL` overrides).\n\n## Ground rules for the agent\n\n1. **Call tools instead of inventing data** — Do not fabricate article headlines, bias scores, prices, forecasts, option fair values, or meme metadata. Fetch with the appropriate tool, then summarize and cite what came back.\n2. **Minimize duplicate calls** — Batch reasoning: one `get_ticker` per symbol per task unless the user asks for a refresh; reuse prior tool output in the same conversation when still relevant.\n3. **Respect limits** — The public tier includes a modest free query allowance; avoid spamming parallel calls. If a call fails for quota or rate limits, say so and suggest spacing requests or upgrading via [heliumtrades.com/mcp-page](https://heliumtrades.com/mcp-page/).\n4. **Trust the client’s tool list** — The hosted server documents **nine** first-class tools below. If your Cursor build lists a different count or names, follow the live `tools/list` output from the MCP panel.\n\n## Tool selection guide\n\nUse **`search_news`** when the user wants **raw articles** matching keywords, with optional filters (source, category, date window, minimum shares, sort). Good for “what are outlets saying about X?” and evidence gathering.\n\nUse **`search_balanced_news`** when the user wants **multi-outlet synthesis** (left/center/right-style balance) with summaries, takeaways, and tickers—not individual article dumps.\n\nUse **`get_source_bias`** when the user names an **outlet** (e.g. “Fox News”, “CNN”) and wants **institutional bias profiles**, signature phrases, similar outlets, and optional recent article breakdowns.\n\nUse **`get_bias_from_url`** when the user supplies a **single article URL** and wants **per-dimension bias scoring** for that page.\n\nUse **`get_all_source_biases`** when the user wants a **landscape view** (many sources at once, ranked/compare outlets). This can be heavy; confirm they want breadth over depth.\n\nUse **`get_ticker`** for **stocks, ETFs, or crypto**: spot context, bull/bear narrative, forecast-style output, IV rank, volatility context, and **hooks into options strategy** content returned by Helium for that symbol.\n\nUse **`get_option_price`** when the user specifies a **single listed option** (underlying, strike, expiration `YYYY-MM-DD`, call/put) and wants **ML fair value** and **probability ITM** from Helium’s models—not broker quotes.\n\nUse **`get_top_trading_strategies`** when the user asks for **ranked options structures** (short vol vs long vol packs, edge-style ranking). Pass sorting preference when the user cares about odds vs reward/risk vs historical performance.\n\nUse **`search_memes`** for **semantic meme search** (keywords against captions/OCR), engagement counts, and image references—useful for culture/market sentiment adjacent to finance.\n\n## Response style after tool calls\n\n- Lead with **answer**, then **compact bullets** of supporting metrics (bias dimensions, price, forecast range, strategy names, meme engagement).\n- Name **limitations** explicitly: model outputs are not investment advice; news coverage may lag fast markets; meme OCR can be noisy.\n- When mixing news and markets, **sequence** tools logically (e.g. `search_balanced_news` → `get_ticker` for tickers mentioned) instead of redundant overlapping searches.\n\n## Security and privacy\n\nDo not paste **secrets** into tool arguments. Article URLs may contain trackers; prefer canonical article links the user provides. Do not exfiltrate unrelated private repo content into search queries.\n"}