When you tell Claude Code “send an SMS to a Korean phone number,” it doesn’t browse a catalog. It calls search_apis with the intent string, and apitree returns a ranked list. The ranking algorithm has four factors — and understanding them is key to why apitree wins.
Every API in the catalog has a MCP Tool Definition — a description, parameter list, and category tag. When an agent searches, apitree computes vector similarity between the intent and every tool definition. This is why description quality matters more than API count.
An API with 99.9% uptime and 150ms p95 latency ranks higher than one with frequent 502s. apitree continuously monitors all 174 Live APIs and computes a composite health score (0-100, grade A-F). Agents naturally gravitate toward reliable providers.
Lower cost-per-call ranks higher when semantic scores are close. Agents are cost-sensitive — they will choose the $0.001/call option over the $0.01/call option if both do the same thing.
APIs that have been called more (and succeeded more) rank higher. This creates a flywheel: popular APIs get recommended more → get called more → get ranked higher. Newcomers break in through better descriptions or lower prices.
RapidAPI ranks by developer reviews. Postman ranks by collection popularity. Both are human signals. apitree ranks by machine signals — because our primary user is an AI Agent, not a human developer.
This is why we invest in Tool Definition quality. A well-written 50-word description can outrank a competitor with 10x more traffic, because agents read descriptions, not logos.