FrelyMCP

Frely CLI

Website · Docs · Frely · English · 简体中文

Frely CLI (frely-cli) gives agents with a shell, such as Claude Code, Codex and pi, the frely Skill: your agent uses your Frely account and the Store from the terminal. For web ChatGPT and Claude, and to work on another computer, it connects a computer you choose to FrelyMCP, so those clients can use the files, shell commands and processes in the folders you allow. The computer connects out to Frely, so you open no ports.

It is also the command line for your Frely account:

• Agents, Prompts and Skills: install a published Frely Agent as a local Skill, or install Prompts and Skills you bought.
• Device MCP: let remote AI clients work on this computer.
• Local MCP servers: pass MCP servers that only run on this computer (for example on 127.0.0.1) through to remote clients.
• Local model sharing: publish a local Ollama or OpenAI-compatible runtime as your own Frely Provider.

Install

Let your agent set it up

Ask an agent that has a shell (Claude Code, Codex, pi) to install the CLI, sign in and install the Skill:

Install Frely CLI and the frely Skill on this computer. On macOS or Linux run curl -fsSL https://frely.cloud/install.sh | sh; on Windows PowerShell run curl.exe -fsSLo "$env:TEMP\frely-install.ps1" https://frely.cloud/install.ps1; powershell -NoProfile -ExecutionPolicy Bypass -File "$env:TEMP\frely-install.ps1". Ask me for my email, run frely login --email <my email>, ask me for the 6-digit code from the email, then run frely login --code <code>. Then run frely skill install --host <your host: claude-code, codex, pi or generic> and read the Skill it installs. Do not run frely link on until I ask for it.

The frely Skill teaches the agent to use your account and to offer FrelyMCP only when a request needs it. This path does not open this computer to remote clients; frely link on does. More

Install it yourself

https://frely.cloud/install.sh and install.ps1 only install the program, for scripts and agents that need the CLI. Node.js 22 or newer is needed only for npm.

# macOS / Linux

curl -fsSL https://frely.cloud/install.sh | sh

# Windows (PowerShell)

curl.exe -fsSLo "$env:TEMP\frely-install.ps1" https://frely.cloud/install.ps1; powershell -NoProfile -ExecutionPolicy Bypass -File "$env:TEMP\frely-install.ps1"

Or with npm:

npm install --global --ignore-scripts frely-cli@latest

The installers check the SHA-256 and measure GitHub Releases and npmmirror in parallel and download from the faster one first (npmjs is the last fallback; set FRELY_INSTALL_SOURCES to pin an order). GitHub Releases carry each executable as frely- .gz (use gunzip, or 7-Zip/tar on Windows) with a .sha256 of the decompressed file; the installers do this for you, so prefer them. More

Then sign in and install the Skill for your agent host:

frely login --email <you@example.com>   # then: frely login --code <6-digit code>

frely skill install --host

Common commands

All commands and options: command reference. frely help --agent --json prints the full machine-readable contract; add --summary for a short grouped index, or a command id or group (for example agent.run, agent) for one command family with examples.

Use with FrelyMCP

Use FrelyMCP from clients with no shell, such as web ChatGPT and Claude, or to work on another computer.

FrelyMCP is one MCP address for your whole account: https://mcp.frely.cloud/mcp. Every computer with remote control on (frely link on) appears behind it.

1. Turn on remote control on the computer

If frely is installed, run frely link on. Otherwise the start scripts below install the latest frely and run frely link on for you; run them again any time to update.

With a browser:

# macOS / Linux

curl -fsSL https://frely.cloud/start.sh | sh

# Windows (PowerShell)

curl.exe -fsSLo "$env:TEMP\frely-start.ps1" https://frely.cloud/start.ps1; powershell -NoProfile -ExecutionPolicy Bypass -File "$env:TEMP\frely-start.ps1"

The browser opens to sign in (new accounts can register there) and to approve this computer, then the MCP URL is printed. The workspace is ~/Frely by default (created for you); set FRELY_WORKSPACE=/path to share another folder. When an AI client needs a folder outside it, you can allow that folder on the web. More

On a server (no browser): get a one-time install command on the Frely website (My Devices → Add a computer → Create install command), or ask ChatGPT or Claude connected to FrelyMCP to add a computer. Paste it on the server:

curl -fsSL https://frely.cloud/start.sh | FRELY_INSTALL_TOKEN=fit_... sh

It signs in to your account and turns remote control on without a browser. The token works once and expires after 15 minutes. On Linux, run sudo loginctl enable-linger $USER so the service stays online after you log out of SSH. More

2. Add FrelyMCP to your client

1. Add https://mcp.frely.cloud/mcp as a remote MCP server with OAuth:

• ChatGPT or Claude: add a custom connector with the URL.
• Claude Code on another computer: claude mcp add --transport http frely "https://mcp.frely.cloud/mcp", then open /mcp to authorize.

1. Sign in to Frely and approve the client.

To check the connection, ask the client to list the top-level names in your workspace. To stop it, run frely link off on that computer.

• What the AI can do: list_devices, then read, edit_file, process and web_fetch on the computer you pick, plus your account through frely_list and frely_call. More
• Local MCP servers: loopback servers are found automatically (frely link local list); each stays hidden until you turn it on in Frely. More
• Authorization: access lasts 90 days by default; frely link on --days 1..365 renews early. The URL never changes. More

More features

• Agents on this computer: frely agent local add downloads an Agent's repository into its own folder and frely agent local start runs it inside the Frely sandbox. It can write only that folder, cannot read your SSH or cloud credentials, and never sees your Frely API key; its model calls go through a local proxy that adds the key. It can reach Frely's API and, after you approve each one once, the sites its frely.yaml lists. To let your own FrelyMCP call one that serves A2A on a loopback port, use frely service share --type a2a --port --label agent- . Details
• Tasks on this computer: frely task add [--model ] -- defines a command that runs inside the Frely sandbox; frely task start --wait runs it in its own worker that survives a closed terminal or a reconnect, and --cron runs it on a schedule through the background service, and frely task share offers it as an A2A service. Details
• Frely-hosted Agents: frely item install --host claude-code installs an Agent as a local Skill; frely agent run calls it. More
• Marketplace Prompts and Skills: frely item install --host claude-code (the same command installs Agents too; add @3 for version 3). Paid parts install only when your account holds a pass. frely item update lists installed items that have a newer version; it changes nothing until you run frely item update . More
• Local model sharing: frely service share model ollama publishes a local model server's models and prints the AccessPoint made for each one (price 0 until you set it; put it in a Plan to sell it). A model endpoint on your private network works too (--url http://192.168.1.20:3000/v1 --key-stdin; you confirm the address in a terminal). To use a model only from tasks on this computer, register it with frely provider add instead. More
• Share a local web page: frely share 3000 asks first, then prints a private link to http://127.0.0.1:3000 through Frely's relay until you press Ctrl+C or it expires (at most 24 hours). Relay traffic is a plan allowance per UTC week; direct connections are not counted. Not the same as frely service share model (local models) or frely tunnel (inside the command sandbox).
• Your account from the terminal: frely api list|describe|call. More

Safety

File tools stay inside the shared workspaces and protect credential and shell start-up files. Commands run as your OS user inside an OS sandbox, and are refused where the computer cannot sandbox them. Clients hold OAuth tokens; the CLI keeps its own key in the system credential store. Is it safe? · Sandbox and permissions · Credential storage

Documentation

• Help: cli.frely.cloud/docs (update, doctor, uninstall)
• Contracts, for contributors: Device MCP · [Device Relay transport](docs/devi