MCP analytics.
How to wire any MCP-speaking LLM (Claude Code, Cursor, Codex) to Eyepup's visitor analytics. The Eyepup MCP server ships as eyepup-mcp on npm. Five tools, one env var, zero config.
Install — Claude Code plugin (recommended)
Bundles the MCP server, a companion subagent that knows when to call which tool, and a skill. One install, full agentic loop.
Install — Claude Code · remote MCP only
Just the five tools, no companion subagent. One URL, no npx, no shell.
Install — Claude Code · stdio
For CI, self-hosted runners, or air-gapped environments where you can't hit a remote URL.
Install — Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"eyepup": {
"command": "npx",
"args": ["-y", "eyepup-mcp"],
"env": { "EYEPUP_API_TOKEN": "epk_live_..." }
}
}
}Install — Codex / OpenClaw / Hermes / any MCP client
Same shape — register a stdio MCP server with command npx -y eyepup-mcp and an EYEPUP_API_TOKEN env var. Every MCP-speaking client supports this configuration.
Five tools your agent can call
eyepup_todo— top friction patterns ranked by impact, with paste-ready promptseyepup_ask— free-text Q&A routed across visitor / funnel / anomaly bucketseyepup_visitor— full dossier for a distinct_ideyepup_visitors_hot— top high-intent visitors right noweyepup_log— write-back changelog so the agent reasons about deploys
stdio today; remote HTTP transport at mcp.eyepup.com with OAuth 2.1 + DCR ships next week — same tools, no env-var handling.
