OceanToken

OceanToken plugins for coding agents

OceanToken is one account for 500+ AI models: GPT, Claude, Gemini, DeepSeek and Qwen for text; GPT Image, Seedream, Recraft and Gemini for images; Seedance, Veo, Sora and Wan for video; plus text-to-speech and transcription. This plugin brings all of it into your coding agent: one command connects Codex, Claude Code, Cursor, Gemini CLI, OpenClaw, OpenCode, VS Code, WorkBuddy, Hermes Agent and Windsurf, and any other MCP client can add the server by hand. Your agent can find a model, check what it costs, generate the asset, and save it into your project.

What your agent gets

Tool	What it does
search_models	Search the catalog by name, type, publisher, capability and price.
get_model	Pricing, limits, and the sizes, durations and inputs a model accepts.
estimate_cost	What a request will cost before it runs, and whether your balance covers it.
generate_image	Text-to-image and image edits, with previews and download links.
create_video + get_job	Video from text, from a first (and last) frame, or from reference media.
generate_speech	Voice-overs and narration.
transcribe_audio	Transcripts and SRT/VTT subtitles.
ask_model	Hand a sub-task to another model: a cheaper one, or a second opinion.
upload_file	Use local files as references.
get_account	Balance, available balance and spend.

Three skills teach the agent how to use them well:

• oceantoken-media: produce images, video and audio for a project.
• oceantoken-models: choose, price and delegate.
• oceantoken-setup: run Codex or Claude Code itself on OceanToken, or wire up your app.

The same plugin folder carries manifests for each host (.codex-plugin, .claude-plugin, .cursor-plugin, and the open Agent Plugins plugin.json), so every host gets the same tools and skills.

Install

You need an OceanToken account (sign up) with some credit.

One command, any agent

Ask your agent:

Install OceanToken for me by following https://mcp.oceantoken.ai/install

or run it yourself:

npx -y @oceantoken/cli@latest connect <client>

<client> is one of codex, claude-code, cursor, gemini, openclaw, opencode, vscode, workbuddy, hermes or windsurf. It installs the plugin or MCP entry and the three skills (Windsurf / Devin Desktop has no skills folder, so it gets the server only), and starts the sign-in. See cli/ for options (API keys, dry run, disconnect, and use to run Codex or Claude Code itself on OceanToken models) and the guide for a walkthrough. The sections below do the same by hand.

Codex

codex plugin marketplace add OceanTokenAI/oceantoken-plugins
codex plugin add oceantoken@oceantoken
codex mcp login oceantoken      # opens the OceanToken sign-in page

In the Codex app you can also add the marketplace, then install OceanToken from the Plugins panel and sign in when asked.

Claude Code

/plugin marketplace add OceanTokenAI/oceantoken-plugins
/plugin install oceantoken@oceantoken
/mcp                             # pick "plugin:oceantoken:oceantoken" and authenticate

Cursor

The button adds the MCP server; Cursor asks you to sign in the first time a tool runs. To get the skills as well, install the plugin from this repository (it has a .cursor-plugin marketplace manifest).

Gemini CLI

gemini extensions install https://github.com/OceanTokenAI/oceantoken-plugins

Then, inside Gemini CLI, run /mcp auth oceantoken and sign in. The extension brings the MCP server, the three skills and a short GEMINI.md.

OpenClaw

openclaw plugins install clawhub:@oceantoken/oceantoken --accept-capabilities
openclaw mcp set oceantoken '{"url":"https://mcp.oceantoken.ai/mcp","transport":"streamable-http","auth":"oauth"}'
openclaw mcp login oceantoken

The ClawHub package brings the three skills (also on ClawHub on their own: oceantoken-media, oceantoken-models, oceantoken-setup). The mcp set entry registers the server with OpenClaw so mcp login can sign you in; it replaces the plugin's own unauthenticated entry of the same name.

Any other MCP client

OceanToken is in the official MCP Registry as ai.oceantoken/mcp, so clients that browse the registry (VS Code and GitHub Copilot, for example) can find it by name.

The server is https://mcp.oceantoken.ai/mcp (streamable HTTP). It supports OAuth with dynamic client registration, so most clients just need the URL. To use an API key instead, send it as a bearer token:

# Codex, without the plugin
export OCEANTOKEN_API_KEY=sk-...
codex mcp add oceantoken --url https://mcp.oceantoken.ai/mcp --bearer-token-env-var OCEANTOKEN_API_KEY

# Claude Code, without the plugin. The single quotes keep the key off the command line:
# Claude Code reads OCEANTOKEN_API_KEY from its environment when it starts.
claude mcp add --transport http oceantoken https://mcp.oceantoken.ai/mcp \
  --header 'Authorization: Bearer ${OCEANTOKEN_API_KEY}'

{
  "mcpServers": {
    "oceantoken": { "type": "http", "url": "https://mcp.oceantoken.ai/mcp" }
  }
}

Muse Code (Meta)

Add the server to the mcp_servers block of your Muse Code settings, then sign in:

{
  "mcp_servers": {
    "oceantoken": { "transport": "streamable_http", "url": "https://mcp.oceantoken.ai/mcp" }
  }
}

muse mcp login oceantoken

Muse Code also reads skills from ~/.agents/skills: copy the folders in plugins/oceantoken/skills/ there to get the OceanToken workflows.

Make your agent reach for OceanToken

Agents choose their own tools. They use OceanToken for anything they cannot do themselves (video, voice-overs, subtitles, models they don't have, cost estimates). For images, Codex also has a built-in generator, so ask for OceanToken by name, or make it a rule:

• Ask explicitly: $oceantoken-media ... in Codex, /oceantoken:oceantoken-media ... in Claude Code, or simply "use OceanToken to ...".

• Make it a project rule in AGENTS.md (Codex) or CLAUDE.md (Claude Code):

  Generate images, video, voice-overs and subtitles with the OceanToken tools.
  Estimate the cost first and ask me before spending more than $2.
  

• Skip approval prompts for the read-only tools in Codex (~/.codex/config.toml):

  [plugins."oceantoken@oceantoken".mcp_servers.oceantoken.tools.search_models]
  approval_mode = "approve"
  [plugins."oceantoken@oceantoken".mcp_servers.oceantoken.tools.estimate_cost]
  approval_mode = "approve"
  [plugins."oceantoken@oceantoken".mcp_servers.oceantoken.tools.get_job]
  approval_mode = "approve"
  

Signing in

The sign-in page offers two ways to connect:

• Sign in: your OceanToken email and password. This creates an API key for the connecting app, which you can see and revoke under API keys in the console.
• API key: paste a key you already have. Give it a credit limit if you want a cap for the agent.

Revoking the key disconnects the app immediately.

Costs

You pay the model's usage from your prepaid OceanToken balance, at the prices on the Models page. Ask your agent for an estimate first (estimate_cost). Video jobs place a small hold when they start and charge the actual cost when they finish. Failed jobs are not charged.

Privacy and terms

Prompts and files go to the model you choose, through OceanToken, to produce your result. Generated files are stored in your OceanToken storage and shared as time-limited links. Besides the MCP server at https://mcp.oceantoken.ai/mcp, the setup skill calls OceanToken's own API at https://api.oceantoken.ai with your OCEANTOKEN_API_KEY, only to check the key and to configure your tools to use it. No other service receives data from this plugin. See the privacy policy and terms.

Support

Open an issue, or email [email protected].