A new open-source project called Teahose MCP is giving AI agents real-time visibility into the startup landscape. The Model Context Protocol server connects Claude directly to a proprietary intel graph covering over 3,000 AI companies, surfacing funding rounds, podcast mentions, newsletter buzz, and emerging market themes on demand. Installation is a single command: claude mcp add teahose -- npx -y teahose-mcp. No signup required to start poking around.
What the Tools Actually Do
The server exposes seven distinct tools for querying the Teahose graph. find_companies runs vector similarity search—pass a competitor URL or plain-text description and get back ranked lookalikes with confidence scores. who_is_talking_about tracks which operators, VCs, and newsletters are mentioning a specific company across podcasts and newsletters, with source links and episode data included. latest_funding pulls fresh rounds with amount, stage, investors, and dates, filterable by theme slug for niche queries like 'humanoid robotics' or 'AI coding assistants'. The remaining tools handle portfolio operations at scale: check_companies batch-queries a watchlist to surface signals from the last 7 or 30 days—useful for keeping CRM entries fresh without manual monitoring. emerging_themes surfaces machine-discovered market themes ranked by recent signal volume, so you catch category shifts before they hit mainstream tech press. get_company returns full profiles with sector tags, theme associations, and aggregated signals across funding, product launches, hiring activity, and media mentions.
Security posture
The project makes deliberate tradeoffs for trust-constrained environments. It's a thin read-only HTTP client that only contacts https://www.teahose.com/api/mcp/v1/*. No filesystem access, no shell execution, no child processes, and critically—zero telemetry beyond the API calls themselves. Only two runtime dependencies: @modelcontextprotocol/sdk and Zod for schema validation. The package.json contains zero install hooks or postinstall scripts, which is rare enough to be noteworthy in the MCP ecosystem.
Rate limits and the free key
Unauthenticated usage caps at 10 requests per minute and 50 per day, returning limited signal windows. A free Teahose API key—available at teahose.com/mcp—raises limits to 30 req/min and 1,000/day while unlocking full data windows (50 funding rows across a 30-day span vs. the basic tier's 15-row/14-day cap). The catch: signing up for a key subscribes you to Teahose's daily digest email, though unsubscribing remains an option.
What this signals for AI agent tooling
Teahose exemplifies a growing pattern in the MCP ecosystem: domain-specific servers that give agents access to proprietary data markets without requiring developers to build connectors from scratch. The company apparently maintains its intel graph independently and exposes it through the MCP interface, effectively monetizing curated startup data while offering the protocol layer as open-source. For builders experimenting with AI-powered deal sourcing, competitive analysis, or market monitoring workflows, this removes a significant friction point—no more stitching together multiple APIs or scraping Crunchbase manually.
The Bottom Line
Teahose MCP is exactly the kind of infrastructure the agent ecosystem needs right now: focused, minimal, and opinionated about what it does. If you're building AI tools that need startup intelligence without inheriting the maintenance burden of a full data pipeline, this is worth bookmarking.