eaglisted
Install: claude install-skill tone-lotto/easy-agnostic
# eag — one source, many agents
`eag` keeps MCP servers in sync from one file, `~/.agents/mcp.json` (user scope) or `./.mcp.json`
(project scope), in the plain `mcpServers` shape. Claude Code and Pi read the project file natively;
Codex gets a rendered copy in `config.toml`; Claude Code user scope gets it through `claude mcp`.
It never overwrites a hand edit, and secrets live in the OS keychain as `${NAME}` references.
Run `eag <command> --help` for every flag. Exit codes everywhere:
`0` nothing to do or done · `1` error · `2` drift (something to apply) · `3` conflict (a native edit eag will not overwrite).
## The commands you actually use
| task | command |
|---|---|
| see what differs between the source and each agent | `eag status --exit-code` (add `--json` in scripts) |
| write the source into every agent | `eag apply` (`--dry-run` first; `--json` in scripts) |
| add a remote server | `eag mcp add NAME --url URL --header 'Authorization: Bearer ${TOKEN}'` |
| add a local server | `eag mcp add NAME --command BIN --arg a --arg b --env K=${V}` |
| store the value behind `${TOKEN}` | `printf %s "$TOKEN" \| eag secret set TOKEN` |
| remove a server everywhere | `eag mcp rm NAME` then `eag apply` |
| keep a server out of one agent | `eag mcp target NAME codex off` |
| make a repo's servers work in every agent | `cd repo && eag init --project && eag adopt claude --scope project && eag apply --scope project` |
| make every repo agnostic at once | `eag adopt claude --all-projec