Connecting an MCP server means handing an agent the right to act on your behalf. We handshake with all 58 955 servers in the registry every 15 minutes and read the tool list each one exposes — so the numbers below are measured, not estimated. 32 921 of them answer with no credentials at all, and 7 697 have lost the repository they were built from while still accepting connections.
Search free React, Tailwind, and shadcn components, blocks, and dashboards; get installable source.
Live tennis scores, players, rankings, odds and win-probability. ATP, WTA, Challenger, ITF, juniors.
Apple developer docs and WWDC transcripts with RAG retrieval and Jev relevance ranking.
Persistent memory for AI agents, for code, research and everyday life. Stale facts retire.
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
Neuronto Agentic Resource Discovery ARD Index. Search every ARD registry + 31,411 verified tools.
Agent Resource Discovery (ARD, IETF draft) registry: search every ARD registry + verified MCP tools
Shopify store and ad intelligence: traffic, best sellers, apps, live Meta ads and weekly breakouts.
Brazilian Open Finance MCP — 30+ banks (Itaú, Nubank, etc.) to Claude/Cursor. Read-only.
Live data from 80+ official national company registries. Unmodified. For KYB and due diligence.
Brazilian legal stack in one MCP: lawsuits, court publications, case law, tenders, certificates.
Conta Azul ERP MCP — sales, customers, finance and NF-e via OAuth 2.0. Read + write, 35 tools.
The description is marketing; the tool list is the contract. Every server page here shows the tools we read straight from the server, with the date we read them.
7 697 servers in the registry point at repositories that are gone. If you cannot open the code, you are trusting a binary you did not build.
A server holding private data should demand a key. 32 921 of them accept anonymous connections — fine for a public API, alarming for anything else.
Not your personal token. A separate account with the narrowest scope that still works means a compromised server costs you one integration, not everything.
Tools get added after you connect. We keep the history of what each server exposed and when, so a server that quietly grew a delete tool last month is visible.
Answers from our own measurements, not from the protocol docs.