Ask your agent to connect
Copy a short, safe prompt for ChatGPT, Claude, Codex, or Claude Code.
Your agent can guide most of the connection setup. A person still completes protected sign-in, and an organisation owner may need to enable the connection before you begin.
Choose the prompt that matches where you work.
ChatGPT or Claude
Use this when you want to connect a hosted assistant. Replace the details you
know and leave the rest as unknown.
Help me connect this assistant to my organisation's Twin using the official
guide at https://docs.mlx.systems/docs/connect.
What I know:
- Organisation or company: <name>
- Twin workspace address: <address-or-unknown>
- Twin gateway address: <address-or-unknown>
- Connection pack: <available-or-not-available>
Please:
1. Identify whether this assistant has an organisation-approved Twin connection.
2. If it does, guide me to add or enable it and let me complete the protected
browser sign-in.
3. If an owner or administrator must act first, stop and tell me exactly what
they need to enable.
4. Never ask me to paste a password, token, client secret, cookie, or API key
into this conversation.
5. Request only the access needed for approved Twin questions.
6. Verify the finished connection with one real, permitted question. Show the
answer and any permitted channel links, and retain the query ID for support.
7. Do not claim the connection is ready merely because its settings were saved.
Do not approve business meaning, publish knowledge, or activate a version on
my behalf.The assistant may be unable to change its own workspace settings. That is normal: it should take you to the correct protected screen or explain which owner needs to act.
Codex or Claude Code
Use this when a local coding agent can run the Twin CLI and open browser sign-in.
Connect this coding agent to my organisation's Twin. Read the live connection
and CLI guides before setup:
- https://docs.mlx.systems/docs/connect
- https://docs.mlx.systems/docs/connect/cli
Use the live guides if installation, sign-in, or a command is unclear. After
installation, treat `twin help --json` as the exact contract for that installed
CLI version.
Connection details:
- Organisation or company: <name>
- Connection pack: <available-or-not-available>
Please:
1. Follow the live Twin CLI guide linked above.
2. Run `command -v twin`. If it is missing, install the public GitHub Release
globally for my user using the exact command in the CLI guide. Do not request
a GitHub account or token, use `npx`, add a project dependency, use `sudo`,
or install a similarly named package.
3. Verify `command -v twin` and `twin help --json` work outside this repository.
If the global npm binary directory is not on my persistent user `PATH`, fix
that without overwriting unrelated shell configuration and tell me to
restart this coding agent.
4. Run `twin login` and let me complete protected browser sign-in.
5. Never ask me for SSH, Tailscale, a tenant or organisation ID, or to paste a token,
secret, password, cookie, or API key into chat, a command, a tracked file,
or a log.
6. Run `twin status --json` and `twin organisations --json`, tell me the current
organisation, and ask me which one to use if several are available. Switch
with `twin organisations --use` without another login, then confirm it with
`twin status --json`.
7. Run `twin channels --json` for the selected organisation.
8. Inspect the relevant channel, then verify the connection with one real,
read-only `twin query --stream --client-app <claude-code-or-codex>` through
the canonical router. Explain
`queued` and `running` status while it works, then present the terminal result.
9. Read channel feeds and flat threads in small pages when useful. List channel
members before using canonical mention IDs. Add a channel post, comment,
reaction, record, view update, or artifact creation/update only after I
approve the exact mutation and content. Use `--allow-write`; use `eyes` for
seen and `check` for handled, and never infer, broaden, or silently retry a
write. Pass the matching `--client-app` value on queries, posts, and
artifacts so MLX can display the initiating client beside my user avatar.
10. If I provide or confirm feedback, submit it with `twin feedback` and attach
the routed query ID when relevant. Never infer feedback on my behalf.
11. Do not configure MCP unless I explicitly ask for it.
12. Report what changed and any step that still belongs to
an organisation owner.
Do not guess internal products or relations, broaden access,
or substitute an unofficial package or connection route.The agent should use the user-scoped CLI unless you explicitly request another supported route.
If you do not have a connection pack
A connection pack tells you:
- which Twin workspace and gateway belong to your organisation;
- which hosted or local tools have been enabled;
- how sign-in maps you to the correct organisation;
- which channels are available; and
- who can help when an owner-controlled step is required.
Ask your organisation administrator or implementation contact for it. Do not guess gateway addresses, organisation identifiers, package sources, or OAuth settings.
A prompt cannot grant itself access
An agent can inspect, configure, and explain. It cannot make an unavailable connection available, approve its own access, or complete a person-owned business review.
Follow the tool-specific guide
ChatGPT
Add the approved app and complete personal or workspace sign-in.
Claude
Add a custom connector or ask a Team or Enterprise owner to enable it.
Codex
Use the Twin CLI first, with optional user-scoped MCP configuration.
Claude Code
Use the Twin CLI first, with optional remote MCP configuration.
Implementation teams preparing the organisation itself should use the operator agent prompt instead.