Indexing the MCP ecosystem: 9 sources, 80,000 servers, and a bunch of nuances

A year and a half ago there were only dozens of MCP servers, now there are tens of thousands, making it hard to find the right one. The author is building Unyly, an aggregator that currently lists more than 80,000 entries from nine sources: the official MCP registry, GitHub, npm, PyPI, crates.io, Docker MCP Catalog, Smithery, Glama, and PulseMCP. Each source has peculiarities: GitHub Search API caps at 1000 results per query so they slice by star ranges; PyPI has no search API so they download a 42 MB simple-index and filter for 'mcp'; PulseMCP randomly returns 410 Gone for half of requests, handled with retries and backoff. They deduplicate by source and slug, giving priority to curated records, and filter out about 30% of packages that are not actually servers. They never guess installation commands, but resolve real ones from repositories; if unknown, the page links to the repo. Each server undergoes a static security scan to produce a score shown before installation. Remote MCP endpoints are health-checked hourly using a real JSON-RPC initialize request. A gateway at gateway.unyly.org/mcp exposes the whole catalog to agents via one remote server with OAuth 2.1, offering three meta-tools: search_mcps, list_mcp_tools, and use_mcp_tool. Live statistics and a directory map are published at unyly.org/stats and unyly.org/mcp-directories. The catalog is free with no paid placements; server owners can submit via unyly.org/submit.