14,920 remote MCP servers, each one live-probed and graded for prompt injection and risky tools. Connect in one click from Claude, ChatGPT or Cursor.
Paste your endpoint: gate scans it in 30 seconds and grades it A–F. At B or better, a short form and a human review put it in the directory with a live tool list, a security badge and one-click connect.
The same scanner behind every grade, plus a control plane for your AI: allow, ask or block each tool, monitor servers for drift, and log every call with SIEM and compliance export.
A server that speaks the Model Context Protocol — the open standard AI clients like Claude, ChatGPT and Cursor use to call external tools. A remote MCP server runs in the cloud, so there is nothing to install: you connect to its URL.
gate connects to every open server, lists its tools and scans them for prompt injection, hidden instructions, exfiltration hints and over-broad permissions, plus a Claude review of each description. 100 points minus deductions; A ≥ 90, B ≥ 78. Grades are re-checked weekly.
A curated set of official vendor endpoints, servers their authors submitted (scanned first, reviewed by a human), and the official MCP registry, which gate imports daily and probes for reachability.
Open the server's page: copy the config for Claude Code, Cursor or ChatGPT, or click Connect for one gateway URL that carries every server you pick, with sign-ins handled for you.
Paste your endpoint on the submit page. If it grades B or better you fill in a short form, a human reviews it, and you get an email either way. Free.
Beyond the directory: a control plane that governs which tools your AI may call (allow / ask / block), monitors servers for drift, and logs every call with SIEM and compliance export.
Connect gate once in Claude, ChatGPT or Cursor. Free to start.
Get started freeA server we're missing, a feature, a client you use — one line is enough. It opens a draft in your mail app; nothing is sent until you hit send.