Skip to main content
Plugins can package remote MCP configuration with agent guidance. Each host uses its own installation and authorization mechanism.

​
ChatGPT custom MCP plugin

Follow ChatGPT setup to add a custom MCP server:
Choose OAuth, approve browser login, then enable EvoLink in your conversation. ChatGPT does not use the Codex marketplace commands below or require a local npm CLI installation. A custom connection does not mean EvoLink is listed in the public plugin directory. Creation and generation-tool access depend on your account, host version, and workspace permissions. See OpenAI custom MCP guidance.

​
Codex repository plugin

Confirm that your Codex version supports plugin. This plugin uses native remote MCP; the website’s CLI prompt is another method. CLI users do not also need a plugin for generation.
1

Add the source and install the plugin

Run in the machine and user terminal that runs Codex:
Confirm EvoLink appears in the plugin list. It provides remote MCP configuration and the evolink-mcp skill, not the EvoLink CLI.
2

Load the plugin and authorize

Refresh plugins or open a new Codex conversation and confirm the EvoLink server is loaded. Complete the host’s browser OAuth when prompted. With the server registered and this command supported, use:
Keep it running through sign-in and approval, then return to Codex. Choose a single registration source when manual MCP configuration already exists.
3

Verify tools and the skill

Success means the skill was read and actual check_balance and search_models calls succeeded; a plugin list alone does not verify authorization.
To update the source, run this and refresh the plugin in the host, checking its actual version:
See Codex CLI reference and Codex setup for manual MCP setup; find CLI instructions in the separate CLI guide.

​
Claude Code repository plugin

Use a plugin-capable Claude Code version. These are Claude Code conversation commands, not normal shell commands.
1

Install the repository plugin

Wait for installation and confirm EvoLink is enabled for the current user/project. Preserve other plugins.
2

Authorize remote MCP

After refreshing the plugin, select EvoLink and complete browser authentication. Avoid a duplicate manually registered connection.
3

Verify for free

Ask the assistant to confirm reading evolink-mcp and actually query balance and models without generation. See Claude Code setup for manual MCP setup; find CLI instructions in the separate CLI guide.

​
Install only the MCP skill

If native MCP is already configured and you only need agent guidance:
This retrieves repository skills through the skills installer. It does not install EvoLink CLI, start local MCP, or complete OAuth. Verify the host discovers the destination and reads evolink-mcp.

​
What the plugin teaches

The source is Evolink-AI/mcp-plugin. Configuration points to hosted remote MCP. Its skill covers model discovery, parameter validation, estimates and approval, reference uploads, generation, tasks, failure recovery, and original delivery. It does not add code development, arbitrary file editing, or full video editing to EvoLink MCP tools. CLI and MCP skills can share media policies, but use different commands and tools. Choose one path per task. If submission times out with an unknown outcome, recover the original request or task_id instead of regenerating through another entry. See agent resources for contents and directories and platform reference for parameters. See costs and retention for limits and revocation. Plugin installation, skill installation, and OAuth authorization do not approve a paid generation request.

​
Plugin installation problems

Plugin installation does not approve paid tests. See central troubleshooting for discovery and other problems.