agent-sync-add-mcplisted

# Add the MCP server to the library, not to this agent's config Every agent stores MCP servers in its own format and its own file — Claude in `~/.claude.json`, Codex in TOML, Cursor in its own JSON. agent-sync keeps one canonical definition and translates it into each dialect. **Edit the library; never hand-edit an agent's config**, or agent-sync will report it as drift and the other agents and devices will never see it. ## Never put a credential in the command line This is the rule that matters most. Tokens, API keys and passwords must not appear in argv — they end up in shell history, and anything in the library is committed to git. **You cannot do this step for the user.** Ask *them* to run: ``` agent-sync secret set <name> ``` It prompts for the value and stores it on that device only, outside the library. Then reference it in the server definition as `${secret:<name>}`, which is what gets committed. For a value that is already an environment variable on every machine, use `${env:VAR}` instead. `agent-sync add mcp` refuses a literal credential rather than committing one. ## The flow 1. **Find out what kind of server it is.** Remote servers have a URL; local ones have a command to run. Ask if it is not clear from what the user gave you. 2. **Add it.** Remote: ``` agent-sync add mcp <id> --url https://example.com/mcp --json ``` Local, with a credential the user has already stored: ``` agent-sync add mcp <id> --command npx --args -y some-