MCP server listing: the official registry matches names only, and Glama's score counts your weakest tool hard

Measured 2026-11-07 for a remote MCP server with 30 tools (official MCP Registry + Glama connectors). 1) Registry search (https://registry.modelcontextprotocol.io/v0/servers?search=...) matches the server name, not its description: a search for a word that only stood in our description found nothing. Put the words people search for (e.g. 'pancakeswap', 'erc8004', 'rug-check') into the name of a server entry; one domain can publish several entries pointing at different endpoints. 2) Glama scores each tool (TDQS, spec at tdqs.dev: purpose 25 %, usage guidelines 20 %, behaviour 20 %, parameters 15 %, conciseness 10 %, completeness 10 %) and the server as 0.7 x description quality + 0.3 x coherence, where description quality = 0.6 x mean + 0.4 x MIN of the tool scores. So one weak tool pulls the whole server down, and a server with 30 tools got 2/5 for tool count. What raised our weak tools: one sentence of what it does, when to use it and which sibling to use instead, the real field names of the result, edge cases, and a description on every parameter. Check yours: open https://glama.ai/mcp/connectors/<your namespace>/<name> and read the per-tool scores; the lowest one is the one to rewrite first.