Skip to main content

​
CLI installation, sign-in, and verification

The CLI supports macOS, Windows, and Linux with Node.js 22+. Install in the machine and user environment where your agent executes commands. Terminal users can follow the manual steps. Paste this entire prompt into the agent’s chat box. It matches the website prompt. Complete browser sign-in and permission approvals yourself.
If the agent cannot run terminal commands, follow the manual steps below. Run bash blocks in a terminal, one step at a time. You do not need to complete both installation methods.

​
Manual installation: five steps

1

Check Node.js and the execution environment

Open a terminal on the machine and under the user that runs the agent. Use Terminal on macOS, PowerShell or Command Prompt on Windows, or Cursor’s integrated terminal.
Success: both commands show versions, with Node.js 22 or newer. If missing or outdated, install a suitable version from Node.js, reopen the terminal, and check again. WSL, SSH hosts, containers, and your desktop are separate environments; install where the agent executes commands.
2

Install and check the CLI

Success: installation finishes without errors and evolink --version shows a CLI version. -g installs a command into the current Node/npm global directory. It does not require cloning an MCP or plugin repository. Fix installation problems before continuing.
3

Sign in to EvoLink

  1. Keep the command running. It opens a browser or prints this attempt’s sign-in link.
  2. Sign in to EvoLink, review the account, client, and requested permissions, and approve.
  3. Return to the original terminal, wait for completion, then run:
Success: authenticated: true. The CLI uses the operating system’s secure credential store; you do not need to create or paste an API Key. A browser callback alone does not complete terminal sign-in. Balance connectivity is checked in the final step. See login troubleshooting for browser, timeout, and Linux storage errors.
4

Install the skill and confirm agent discovery

Without --agent, installation targets all supported skill directories; it does not install those agents. Use an agent value listed below to select one host.Success: current: true. In the agent chat, send:
The skill guides model discovery, parameters, estimates and approval, references, tasks, and original downloads. It does not register native MCP or sign you in. assistant_discovery: not_checked requires confirmation from the host. Refresh skills or open a new conversation if needed. See skill discovery.
5

Verify for free and confirm readiness

Success: balance returns account data; doctor reports ok, connection_verified, and model_discovery_verified as true; the agent confirms reading the skill. Zero balance can still establish connectivity, but generation needs sufficient funds.These steps do not create paid media tasks. Once complete, the assistant can say you are ready. Fix any failed check before proceeding; do not generate an image as a connectivity test.

​
Install the skill for one agent

The website’s evolink skills install targets all supported directories. For one host, replace AGENT_NAME below with a value from this table:
For example, Gemini uses:
These are CLI installation targets, not a guarantee that every host version automatically discovers the directory. The host must permit commands and load skills; see skill and permission troubleshooting. Unlisted assistants can read Agent-readable surfaces; do not invent agent values.

​
Update the CLI and skill

Check current: true and ask the agent to reread the skill. Local edits are preserved. Use --replace-modified only after deciding to replace them; previous content is backed up under ~/.evolink-media/skill-backups/. Do not use the old @evolinkai/media-cli / evolink-media for new installations.

​
Inspect sign-in or sign out

Check the current session:
Run this separately only when you want to sign out:
This revokes the current CLI session. Manage native MCP authorization in its own host. Save existing task IDs before switching accounts; old quotes cannot be submitted as a new account.

​
Linux and SSH: browser on a different machine

Persistent Linux sign-in needs D-Bus and an unlocked Secret Service in the current user session. Check this first. Without it, credential_store_unavailable is expected; root access is not a substitute. Configure secure storage or run the CLI locally in an environment that provides it.
1

Start sign-in on the remote host

Keep this command running and use the current authorization link. --no-browser suppresses browser launch but still waits for a callback; it is not device-code login. The range is 30–900 seconds, default 180.
2

Forward the callback port from a local terminal

Find the actual port in this attempt’s redirect_uri. Replace both CALLBACK_PORT values, plus USER and HOST. For http://127.0.0.1:54321/callback, use 54321 in both positions. Do not publish the full authorization link.
Keep the forwarding terminal running; an idle terminal without a prompt is normal. Resolve a port conflict without editing the authorization link’s port.
3

Open this link and verify remotely

Open the first step’s link in your local browser, sign in and approve, then wait for the remote command to finish. In that same remote user environment, run:
Stop this forwarding session after success. On timeout, start a new login and forward its actual port; the old link cannot be reused. Native MCP callbacks belong to the host and do not use EvoLink CLI flags.

​
After installation

Follow the first generation workflow: describe the task, have the agent check available models, parameters, and cost, then confirm before submission. See the CLI command reference and central troubleshooting for installation, sign-in, permissions, and downloads. For native remote MCP or local stdio, use the other MCP clients guide. CLI sign-in options do not configure an MCP connection.