EvoLink

EvoLink MCP plugin

[smithery badge]

Use 150+ image, video and audio models (Seedance, Kling, Veo, Sora, GPT Image, Nano Banana, Suno and more) from your AI assistant. This repository packages EvoLink's remote MCP server, https://mcp.evolink.ai/mcp, with a skill that teaches the assistant to find a model, quote the price and generate only after you confirm.

• No API key to copy. The first time, you sign in to EvoLink in your browser and approve the connection.
• Pay as you go. Generation is charged to your EvoLink balance at the same prices as the API. The MCP itself is free and needs no subscription.
• Website: evolink.ai/mcp

Install

Claude Code

One command adds EvoLink and signs you in (run it in your terminal, or type ! in Claude Code and paste it):

claude mcp add --transport http --scope user evolink https://mcp.evolink.ai/mcp && claude mcp login evolink

Your browser opens the EvoLink sign-in page; approve, then reopen Claude Code. On Windows PowerShell 5, replace && with ;.

To get the skill as well, install the plugin instead (run /reload-plugins if Claude Code asks you to), then type /mcp, choose evolink and select Authenticate:

/plugin marketplace add Evolink-AI/mcp-plugin/plugin install evolink@evolink

Codex

One command; Codex 0.77 or later opens the browser sign-in by itself:

codex mcp add evolink --url https://mcp.evolink.ai/mcp

Run codex mcp login evolink only if the output says the server "may or may not require login" or no browser opens.

Or install the plugin, which also adds the skill, then sign in with codex mcp login evolink:

codex plugin marketplace add Evolink-AI/mcp-plugincodex plugin add evolink@evolink

After signing in, ask the assistant to call EvoLink’s check_balance in the current conversation. If it succeeds, keep using it. Only if tools are missing, exit the CLI and run codex resume --last; in the Codex app, start a new chat and restart if needed.

To let Codex do the setup, send it this prompt:

Set up EvoLink in Codex for me so I can generate images, videos and audio from here.1. Install the companion skill: run `npx skills add Evolink-AI/mcp-plugin -g -y`. Then run `npx skills ls -g -a codex` to verify evolink-mcp from its output.2. Add the EvoLink MCP server: run `codex mcp add evolink --url https://mcp.evolink.ai/mcp`. It starts the sign-in in my browser; if no browser opens, give me the link it prints.3. I'll sign in to EvoLink in the browser and approve. Then call EvoLink's check_balance in this conversation to verify the connection. Only if its tools are missing, ask me to reopen: exit the CLI and run `codex resume --last`; in the Codex app, start a new chat, then restart the app if needed.After a successful check, tell me it is ready. Use EvoLink by default for media generation; quote each paid task and wait for my explicit approval.

Cursor

Add to Cursor, then find evolink in Cursor's MCP settings and click Connect to sign in. Or add this to ~/.cursor/mcp.json:

{  "mcpServers": {    "evolink": {      "url": "https://mcp.evolink.ai/mcp"    }  }}

With the Cursor CLI, sign in with agent mcp login evolink after adding the config.

Cline and Careti

Native remote setup, browser sign-in, and free verification steps are documented in llms-install.md. Use Streamable HTTP and keep tool approval prompts enabled. Adding the MCP connection is separate from installing the optional companion skill.

Gemini CLI

Install the extension from this repository:

gemini extensions install https://github.com/Evolink-AI/mcp-plugin

Review the extension permissions, then open Gemini CLI and use /mcp to inspect evolink. If authentication is required, run /mcp auth evolink and sign in to EvoLink in the browser. This remote connection needs no EvoLink CLI or API key. If you already configured a server named evolink in your Gemini settings, that configuration takes precedence; check it before adding another connection.

Verify actual free check_balance and search_models calls before generating. Connection and discovery are free; generation uses your EvoLink balance. Review an estimate and approve each paid request. Setup issues: [email protected].

Hermes Agent

hermes mcp add evolink --url https://mcp.evolink.ai/mcp --auth oauth --connect-timeout 300

The browser sign-in opens by itself; press Enter when asked to enable the tools, then /reload-mcp in an open session. (Based on Hermes' documentation; not tested by us yet.)

OpenClaw

Run on the machine that hosts the OpenClaw Gateway, then open the printed link to sign in:

openclaw mcp add evolink --url https://mcp.evolink.ai/mcp --transport streamable-http --auth oauth && openclaw mcp login evolink

(Based on OpenClaw's documentation; not tested by us yet.)

Claude (web and desktop)

Add to Claude opens the Add custom connector dialog with the name and URL filled in. Click Add, then Connect, and sign in to EvoLink. Claude's Free plan can add one custom connector.

Skills only

npx skills add Evolink-AI/mcp-plugin

This installs the evolink-mcp skill for agents that support skills. Add the MCP server in your client as shown above; the skill alone cannot generate anything.

Upgrading from evolink-media

Install the new skill first and check that evolink-mcp appears in your agent. For a standalone global install, remove the old skill with npx skills remove evolink-media -g -y. Back up any personal edits first. For a plugin-managed skill, update the plugin through its client instead. Do not add a second MCP connection; the server name stays evolink.

What the assistant can do

The hosted server can't read paths on your computer. Give reference files as public links, pass files under 1 MB directly, or let an assistant that can run commands (Claude Code, Codex, Cursor) upload a local file of up to 95 MB with prepare_upload. Uploads are free and deleted after 72 hours.

Default provider and confirmation

The evolink-mcp skill instructs the assistant to use EvoLink by default for media generation and editing, including variations and regeneration. It should ask before switching to another provider unless you explicitly requested it.

Before a paid task, the assistant should quote the exact settings and wait for explicit approval. Only your explicit approval of a task or batch with a sufficient budget allows it to continue without asking again; the assistant’s suggested budget is not approval. Partial or unknown costs must be explained before you decide to proceed. New generations outside an approved batch need a new confirmation.

These are assistant instructions, not a server-side approval gate. Client tool approval prompts depend on your settings. In Codex, set approval_mode = "prompt" under each [mcp_servers.evolink.tools.generate_image], [mcp_servers.evolink.tools.generate_video] and [mcp_servers.evolink.tools.generate_audio] entry if you want a tool prompt for every paid call.

Costs and control

• After your first connection, API Keys in the EvoLink dashboard shows a key named EvoLink MCP (OAuth), shared by all your assistants.
• There is no spending limit by default. Set a total limit, a daily limit or an expiry date on that key to cap MCP spending; disable it to pause MCP for every assistant.
• MCP tasks and charges appear in the dashboard like any other usage.

What's in this repository

The MCP server itself runs at https://mcp.evolink.ai/mcp; this repository holds only the client configuration and the skill.

Network endpoints and credentials

• MCP server: https://mcp.evolink.ai/mcp (Streamable HTTP). Every tool call goes there. Prompts, parameters and reference file links are sent to EvoLink to run the task; results come back as links.
• Sign-in: OAuth 2.1 with PKCE through https://passport.evolink.ai, with the approval page on evolink.ai. Scopes are mcp and offline_access. Your assistant stores the tokens it receives.
• No secrets or local code: this repository contains no API keys, hooks, scripts or post-install commands. It holds configuration files and a Markdown skill.

Support

• Setup guide: evolink.ai/mcp
• Email: [email protected]
• Privacy policy: evolink.ai/privacy
• Terms of service: evolink.ai/terms

License

Apache-2.0