Agent Connector

Two audiences. (A) MCP DEVELOPER — write an MCP server, lifecycle hooks (12 normalized events + a native-event passthrough), slash commands, Agent Skills, subagents, or standing AGENTS.md guidance (memory) ONCE with defineConnector({...}), then install/sync/uninstall them across every detected AI-agent CLI (Claude Code, Codex, Cursor, Copilot, Gemini, OpenCode, Warp, and more — 32 registered deploy adapters) in each host's native config dialect, with default local-first per-tool token telemetry for YOUR OWN wrapped stdio server. (B) AGENT-CLI END USER — with NO connector at all, run `agent-connector usage` to see per-CLI / per-model token totals scanned read-only from each agent CLI's own session logs. Use this when a developer wants one integration to reach many agent hosts and to see which of their own server's tools cost the most context, OR when any agent-CLI user wants whole-conversation token totals per CLI/model with zero setup.

mkdir -p ~/.claude/skills/agent-connector
curl -L https://claudskills.com/skills/agent-connector/SKILL.md \
  -o ~/.claude/skills/agent-connector/SKILL.md

@misc{agent-connector-2026,
  author    = {ken-jo},
  title     = {Agent Connector [Claude Code skill]},
  year      = {2026},
  publisher = {ClaudSkills},
  url       = {https://claudskills.com/skills/agent-connector/}
}