Summary

A decision step before integration code, with the evidence rules built into the write path.

The review side is what keeps the data honest: submit_review demands an evidenceLevel, and the API refuses live-success fields on a docs-only report — so a score cannot be inflated by an agent that never made a request. The read side earns its place through get_api_docs, which returns the exact auth header and env var name rather than a link to go read.

What it is

An MCP server over the CLIRank directory, meant to be called before an agent installs an SDK or writes integration code. It answers which API fits a task with present-day pricing, auth method and headless-CLI signals, rather than leaving the choice to model memory or a broad web search.

What you get

discover_apis — search by intent, filtered by minimum CLI relevance score and pricing model (free, freemium, paid, pay-per-use or transaction-based)
recommend — one opinionated pick for a concrete task, with the runner-up and the reasoning
get_api_details — scores, pricing, auth method, capabilities and quality breakdown for one API by slug
get_api_docs — an agent-shaped quickstart: current base URL, the exact auth header, the environment variable name, the SDK install command and an init snippet
compare_apis — two to five APIs side by side on score, pricing, auth and capabilities
browse_categories and get_reviews — the category list with counts, and integration reports from humans and agents
submit_review — a structured report after a real attempt; evidenceLevel is required, and live-success claims are stripped unless the level says a real request was made

Requirements

Nothing — clients run npx -y clirank-mcp-server@latest over stdio and it talks to https://clirank.dev/api with no CLIRank key. There is also a hosted endpoint at https://clirank-mcp.fly.dev/mcp for clients that speak remote MCP. Set CLIRANK_API_URL only to point at a non-production API.

Setup effort

One command — npx -y clirank-mcp-server@latest