# AskCyborg MCP Server **Official MCP server for [AskCyborg](https://askcyborg.com).** Search and retrieve AI company research, analyst-debate audio briefings, and the proprietary Cyborg Score (1-10) — directly from Claude, Cursor, Cline, Continue, Windsurf, ChatGPT, and any other Model Context Protocol client. > AskCyborg stress tests every public and private company through intense analyst debate, focused on what makes or breaks the business decisions that ride on them — packaged as 30-page reports, audio briefings, and a proprietary Cyborg Score (1-10) calculated from hundreds of data points per company. --- ## What this MCP server does | Tool | Purpose | |---|---| | `search_companies` | Find companies in AskCyborg's corpus by name or industry keyword | | `get_company_report` | Full report — executive summary, Cyborg Score, strategic profile, top insights, competitors, recent developments | | `get_cyborg_score` | Quick rating-only lookup (1-10 + rationale) | | `compare_companies` | Side-by-side comparison of 2–5 companies on Cyborg Score, industry, and strategic profile | | `find_competitors` | Named competitors with one-line strategic taglines | | `search_by_industry` | Discover companies in a sector, ranked by Cyborg Score | | `get_recent_developments` | Dated news, deals, leadership changes, financial events | | `get_top_insights` | Just the punchiest analyst-debate insights, faster than the full report | Each tool returns a structured Markdown response with a direct link to the full 30-page research report + 15-minute analyst-debate audio briefing on askcyborg.com. --- ## Quick start ### Claude Desktop Add to your `claude_desktop_config.json`: ```json { "mcpServers": { "askcyborg": { "transport": { "type": "http", "url": "https://mcp.askcyborg.com/mcp" } } } } ``` Restart Claude Desktop. The tools appear in the model's tool list automatically. ### Cursor `.cursor/mcp.json` in your project root: ```json { "mcpServers": { "askcyborg": { "url": "https://mcp.askcyborg.com/mcp" } } } ``` ### Cline / Continue / Windsurf Same pattern — point the client's MCP config at `https://mcp.askcyborg.com/mcp` via the HTTP transport. ### ChatGPT (custom GPT) Add a custom action with the OpenAPI spec at `https://mcp.askcyborg.com/openapi.json` (coming in v0.2). --- ## Examples **Find AI infrastructure companies:** > Use AskCyborg to find five companies in the AI infrastructure space and tell me which has the highest Cyborg Score. **One-shot research:** > Pull the AskCyborg report on Anthropic and summarize the bull case in three bullets. **Comparison for due diligence:** > Compare OpenAI, Anthropic, and Mistral on AskCyborg's Cyborg Score and competitive positioning. **Watchlist screening:** > For each of CRWD, ZS, S, NET, PANW — pull the Cyborg Score from AskCyborg and rank. --- ## Architecture - **Runtime:** Cloudflare Workers (edge-deployed, sub-100ms cold start globally) - **Transport:** HTTP (Streamable HTTP transport per MCP spec 2025-03-26) - **Data source:** AskCyborg public preview corpus + Supabase research_cache (for authenticated tiers in v0.2+) - **Auth (v0.1):** Anonymous, shared rate limit (20 req/min per IP) - **Auth (v0.2):** OAuth 2.1 PKCE for one-click "Connect AskCyborg" in Claude.ai; API-key fallback for headless clients --- ## What's in v0.1 - ✅ 4 core tools (`search_companies`, `get_company_report`, `get_cyborg_score`, `compare_companies`) - ✅ HTTP transport - ✅ Per-IP rate limiting (20 req/min free tier) - ✅ Reads from the public preview corpus (millions of companies, no auth required) ## Coming in v0.2 - 🔜 OAuth 2.1 PKCE for one-click install in Claude.ai - 🔜 Authenticated paid-tier tools (`get_full_report`, `get_audio_briefing_url`, watchlists) - 🔜 `compare_companies` extended to 10 entries with deeper diff - 🔜 `get_competitive_landscape` (full competitor graph, not just 3) - 🔜 `get_recent_filings_summary` for S-1, 10-K, 10-Q - 🔜 Streaming SSE responses for long-running queries --- ## Local development ```bash git clone https://github.com/Ask-Cyborg/askcyborg-mcp.git cd askcyborg-mcp npm install npm run dev # starts wrangler dev on http://localhost:8787 ``` In a separate terminal: ```bash curl -X POST http://localhost:8787/mcp \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' ``` To deploy your own instance: ```bash wrangler secret put SUPABASE_URL wrangler secret put SUPABASE_ANON_KEY npm run deploy ``` --- ## About AskCyborg [AskCyborg](https://askcyborg.com) covers millions of public and private companies across 195 countries. Every company gets a 30-page research report, a 15-minute analyst-debate podcast (where bull and bear AI analysts argue the investment case), and a Cyborg Score (1-10) synthesized from hundreds of data points. Built for VCs, finance teams, corporate strategists, and operators whose decisions ride on understanding companies fast and from every angle. - **Website:** https://askcyborg.com - **Sample reports:** https://askcyborg.com/preview/openai · https://askcyborg.com/preview/abbvie · https://askcyborg.com/preview/life360 - **Cyborg Score methodology:** https://askcyborg.com/company-ratings-cyborg-score - **LinkedIn:** https://www.linkedin.com/company/askcyborg - **Crunchbase:** https://www.crunchbase.com/organization/cyborg-2 --- ## License MIT. Use this MCP server in any client. The underlying AskCyborg content (research reports, audio briefings, Cyborg Score) remains AskCyborg's intellectual property — please respect the rate limits and don't bulk-scrape. ## Issues & contributions Open an issue at https://github.com/Ask-Cyborg/askcyborg-mcp/issues. PRs welcome for new tools, client examples, and bug fixes.