ken-jo/agent-connector

ken-jo

TypeScript8updated 2mo ago

Deploy one MCP to every agent host. Declare your MCP server once with defineConnector() → native config, hooks, skills & plugin bundle across 42 agent hosts (Claude Code, Codex, Cursor, Copilot, Gemini CLI, Windsurf, Zed, Kiro, Warp, OpenCode…). Your users install with one command. Apache-2.0.

ClaudeCodexCursorMCP

agent-pluginsagent-skillsai-agentsclaude-codeclicodexcursordeveloper-toolsgemini-cligithub-copilothooksllm-tools

View on GitHub

README

agent-connector

If you BUILD an MCP integration: write your server + hooks once with defineConnector(), deploy it to every detected AI-agent platform, and measure your own server's per-tool tokens. If you just USE agent CLIs: run agent-connector usage to read their logs and see per-CLI / per-model token totals — no connector, config, or install.

Who this is for

agent-connector serves two distinct audiences — pick your track:

I build an MCP integration (MCP developer) → you write your server + hooks once and deploy them everywhere, then measure your own server's per-tool tokens. Start at Quick start → MCP developer.
I just use agent CLIs and want to see token usage (agent-CLI user) → you already run Claude Code / Codex / Cursor and haven't authored a connector; you just want per-CLI / per-model token totals. Run agent-connector usage — no connector, config, or install required.

The dividing line: the connector-free usage path reports whole-conversation totals per agent CLI / model / project / session / day. It does