Manyfold CLI
Use mf to manage agents, runtimes, channels, files, automations, and more from a terminal.
The mf CLI is the terminal interface to Manyfold. Use it for interactive administration, scripts, CI jobs, and connecting a self-owned computer. It is distributed as a standalone binary for macOS, Linux, and Windows; normal use does not require Node.js.
Start with Install the CLI, then sign in:
mf setup
mf whoami
mf --help
mf setup signs in, registers the current machine, and starts its local
daemon. If you only want the client, install without setup and run mf login.
What you can manage
| Task | Command | Examples |
|---|---|---|
| First-time setup | mf setup | Sign in, register this machine, and start its daemon in one flow. |
| Authentication and identity | mf login, mf whoami, mf auth | Sign in, inspect the current identity, request a missing capability. |
| Agents | mf agent | List, inspect, create, update, delete, configure credentials, and report storage usage. |
| Runtimes | mf runtime | Inspect and manage agent runtimes, hosted framework agents, control UI, and Hermes dashboard. |
| Model configuration | mf model-config | Read/update an agent’s model configuration and refresh its model list. |
| Channels | mf channels | Create, update, register, test, send through, and manage sessions for IM channels. |
| Automations | mf automations | Create schedules, update them, inspect runs, or trigger a run immediately. |
| Runtime files | mf files | List, read, write, move, and remove files exposed by an agent runtime. |
| Backups | mf backups | Create/list backups, restore an agent, and inspect restore status. |
| Skills | mf skills | Discover, install, enable/disable, and uninstall agent skills. |
| Usage | mf usage | Query summaries, time series, events, session totals, and top agents. |
| Connections | mf connections | List connections available to the current agent or account. |
| Agent-to-agent | mf a2a | Inspect Agent Cards, call granted peers, and track A2A tasks. |
| Self-owned computer | mf daemon | Register a machine, install autostart, check status/logs, and diagnose frameworks. |
| Profiles and environments | mf profile | Inspect, select, and remove isolated CLI control-plane profiles. |
| CLI lifecycle and help | mf update, mf help | Check/install CLI updates and open human- or agent-oriented help. |
Run mf <command> --help before a write or destructive operation to see the current arguments and flags. The CLI help is the exact reference for the installed version.
Common workflows
Inspect an agent
mf agent list
mf agent get agt_xxx
mf runtime list
mf model-config get agt_xxx
Work with channels
mf channels list --agent-id agt_xxx
mf channels test chn_xxx
mf channels sessions list chn_xxx --scope-key '<scope>'
See Connect channels for provider setup and Session switching for the chat command model.
Read or write runtime files
mf files roots agt_xxx
mf files list agt_xxx workspace
mf files read agt_xxx workspace/README.md
mf files write agt_xxx workspace/note.txt --content 'hello'
To move whole files, use upload and download. They stream, so transferring a
large file does not depend on its size fitting in memory, and a download that is
interrupted leaves the existing local file untouched:
mf files upload ./report.csv workspace/report.csv --agent-id agt_xxx
mf files download workspace/report.csv ./report.csv --agent-id agt_xxx
With --agent-id set (or MF_AGENT_ID in a runtime), the agent argument can be
left off any mf files command: mf files ls workspace.
Paths must stay inside a file root exposed by the agent runtime. mf files roots
reports each root’s upload and download limits; an upload over the limit is
refused before the transfer starts.
Manage an automation
mf automations list --agent-id agt_xxx
mf automations get aut_xxx
mf automations run aut_xxx
Use mf automations create --help or update --help for the current schedule and payload options.
Connect your own machine
mf daemon register --token ldt_xxx
mf daemon status
mf daemon doctor
Follow Register a self-owned computer for token issuance, autostart, and troubleshooting.
Authentication and context
For normal terminal use, mf login opens browser authentication and stores the resulting CLI profile locally. Use mf whoami to verify which account is active. When the terminal has no browser of its own — an SSH session, for example — add --no-launch-browser to sign in by pasting an authorization code instead. See Install the CLI.
Global options can override the configured context for one command:
| Option or environment variable | Purpose |
|---|---|
--profile <name> / MF_PROFILE | Select an isolated CLI profile. Stable binaries default to default; dev binaries default to staging. |
--api-url <url> / MF_API_URL | Use a different Manyfold API deployment. |
--token <token> / MF_TOKEN | Override the stored credential for one shell or command. Use --token - to read from stdin; direct values may appear in shell history and process lists. |
--agent-id <id> / MF_AGENT_ID | Select an agent context for commands that support it. |
--account | Explicitly operate across the account instead of only the current agent context; may require user-granted permission. |
MF_HTTP_TIMEOUT | Set the timeout for ordinary API requests. The default is 30s; use a plain number for seconds or a duration suffix such as ms, s, m, or h. |
See Profiles and environments before using one machine with multiple Manyfold deployments.
Most resource commands support --json for scripts. Successful payloads go
to stdout; structured failures go to stderr with stable exit codes. See
Scripting with mf for the complete contract.
Agent identities and permissions
Inside a Manyfold-managed agent runtime, mf can use the injected agent identity instead of an interactive login. Operations on that agent are scoped automatically. Account-wide access or another agent’s resources may require an explicit grant.
When an agent reports a missing capability, request only the required scope:
mf auth ensure --scopes channels:read,channels:edit
The command produces a consent URL for the user to approve. Never share or print the underlying token.
Updating and troubleshooting
mf update --check
mf update
mf help
- Installed standalone binaries can self-update on macOS, Linux, and Windows. Downloads are SHA-256 verified and extracted in-process; no system
tarorunzipcommand is required. - Use
mf <command> --helpwhen an option is rejected; commands can change between CLI versions. - Use
mf whoamiwhen authentication or account selection looks wrong. - Use
mf daemon doctorfor self-owned computer and local framework problems. - After updating a running local daemon, restart it so the autostart service uses the new binary.
See also
- Install the CLI
- Profiles and environments
- Scripting with mf
- CLI command reference
- Register a self-owned computer
- Manage agents with the CLI
- Manage runtimes with the CLI
- Manage automations with the CLI
- Back up and restore agents
- Manage skills with the CLI
- Query usage with the CLI
- Call peer agents with the CLI
- Connect channels