OceanToken

What OceanToken MCP server does

The OceanToken MCP server gives MCP-compatible coding agents access to a catalog of more than 500 models through one OceanToken account. The catalog includes models for text, images, video, speech, and transcription. Agents can inspect model capabilities and pricing before choosing a model, then submit generation or delegation requests.

The server also exposes account information and file handling. Generated assets can be saved into a project using the links returned by the service, while local files can be uploaded as references. Use this integration when an agent needs media generation, subtitles, model comparison, or a second model for a sub-task.

How it works

OceanToken runs as a remote streamable-HTTP MCP server at https://mcp.oceantoken.ai/mcp. Authentication normally uses OAuth with dynamic client registration. Clients that do not use the sign-in flow can send an OceanToken API key as a bearer token.

An agent can first search the model catalog, retrieve a model’s supported inputs and limits, and estimate the expected cost. It can then call the relevant generation or delegation tool. Video creation is asynchronous: the agent starts a job and checks its status with get_job. Video requests can use text, an initial and final frame, or reference media.

Setup and configuration

An OceanToken account with credit is required. The repository provides installation paths for Codex, Claude Code, Cursor, Gemini CLI, and OpenClaw, along with a command-line installer for several hosts. Other MCP clients can register the streamable-HTTP URL directly, provided they support OAuth or bearer-token authentication.

For API-key configuration, set OCEANTOKEN_API_KEY and pass it as a bearer token in the client’s MCP configuration. OAuth users generally need only the server URL and the client’s authentication command. Cursor, Gemini CLI, OpenClaw, and other supported hosts have their own sign-in steps documented by the project.

Tools and capabilities

The OceanToken MCP server provides tools for:

• Searching models by name, type, publisher, capability, or price.
• Reading model pricing, limits, accepted inputs, sizes, and durations.
• Estimating request cost and checking whether the account balance is sufficient.
• Generating and editing images, including previews and download links.
• Creating videos and checking asynchronous job status.
• Generating voice-overs and narration.
• Transcribing audio and producing SRT or VTT subtitles.
• Asking another model to handle a sub-task or provide a second opinion.
• Uploading local reference files and reading account balance and spending data.

The repository also includes skills for media production, model selection and delegation, and setup workflows for supported agents.

Limitations and notes

Usage is charged against a prepaid OceanToken balance at the listed model prices. Video jobs place a temporary hold when they begin and charge the final amount after completion; failed jobs are not charged. Prompts and files are sent through OceanToken to the selected model. Generated files are stored in OceanToken storage and exposed through time-limited links.

The OceanToken MCP server does not make every model available for every task; model inputs, limits, sizes, durations, and pricing vary. Agents should use the catalog and cost-estimation tools before starting paid work. API keys can be revoked from the OceanToken console, which disconnects the associated application.