remembers — one command to connect every agent harness to Minds

npx remembers              # sign in with LifeID, pick harnesses, done
npx remembers disconnect   # undo everything connect did (and sign out)
npx remembers status

connect:

1. Signs you in with LifeID through the Minds OAuth server at mcp.minds.sh (which hands off to auth.life.one). It reads /.well-known/oauth-authorization-server, opens a free 127.0.0.1 port, registers that exact /callback at /oauth/register (public client), runs authorize with PKCE S256, and exchanges the code at the token endpoint. The refresh token is stored 0600 in ~/.config/minds/session.json, and later runs refresh silently. Set MINDS_TOKEN for headless/CI.
2. Detects installed harnesses and shows a checklist.
3. Adds one MCP server, https://mcp.minds.sh/mcp, to each, which covers all your Minds (the bridge lists them and every tool takes an optional mind).
4. Asks: Also route this harness's AI calls through Foundry? If you say yes, it mints an org-scoped Foundry key with your LifeID session (stored 0600 in ~/.config/foundry/key, shared with foundry connect) and points the harness at api.fdy.sh (OpenAI and Anthropic formats).

Every file is backed up (*.minds-bak.<timestamp>) before it is written. Re-running replaces only the CLI's own entries, so it's idempotent. No URL contains a secret, and keys never appear in logs.

| Harness | Detected by | Minds MCP written to | Foundry routing | |---|---|---|---| | Claude Code | claude on PATH, ~/.claude* | claude mcp add --transport http -s user (fallback ~/.claude.json) | ~/.claude/settings.json env (Anthropic format); prior values restored on disconnect | | Codex | codex, ~/.codex | ~/.codex/config.toml [mcp_servers.minds] (managed block) | [model_providers.foundry] + profile = "foundry"; needs source ~/.config/foundry/env | | Cursor | ~/.cursor, cursor | ~/.cursor/mcp.json | manual (Settings > Models); CLI prints values | | Grok Bot | ~/.grokbot, grokbot | ~/.grokbot/mcp.json | manual (Model providers); CLI prints values | | OpenClaw | ~/.openclaw, openclaw | ~/.openclaw/mcp.json | source ~/.config/foundry/env (OpenAI env vars) | | Hermes | ~/.hermes, hermes | ~/.hermes/config.yaml mcp_servers.minds (nested under an existing key if present) | source ~/.config/foundry/env | | VS Code / Copilot | Code/User dir, code | <Code/User>/mcp.json servers.minds | manual (Chat > Manage Models) | | Windsurf | ~/.codeium/windsurf, windsurf | ~/.codeium/windsurf/mcp_config.json (serverUrl) | not supported by Windsurf |

Flags: --yes, --only=claude,codex, --foundry / --no-foundry, --no-login, --dry-run. Env overrides (for testing): MINDS_HOME, MINDS_MCP_URL, MINDS_OAUTH_ISSUER, MINDS_NO_BROWSER, LIFEID_TOKEN (Foundry key minting), FOUNDRY_BASE_URL, MINDS_NO_NATIVE_CLI.

Tests: npm test runs the full connect → re-connect → disconnect cycle against fake home dirs for all 8 harnesses, with a mock Foundry key endpoint, and checks that no secret reaches stdout.

Package: remembers on npm (installs the remembers command, with minds as an alias). npx remembers runs connect.