MCP Openai

What mcp-openai does

The mcp-openai MCP server gives compatible clients access to OpenAI operations through AceDataCloud. It covers conversational generation with chat completion and Responses API tools, image creation and editing, text embeddings, speech synthesis, and audio transcription from a URL.

The server also exposes tools for listing available chat, image, and embedding models. A usage-guide tool provides additional information from the service. Supported model families described by the project include GPT-4, GPT-4o, GPT-5, o1, o3, o4-mini, gpt-image-1, dall-e-3, nano-banana, text-embedding-3, whisper-1, and gpt-transcribe, subject to the account and service access available through AceDataCloud.

How it works

Clients communicate with the hosted MCP endpoint at https://openai.mcp.acedata.cloud/mcp, or launch a local stdio process installed from the mcp-openai-pro distribution. In both cases, requests are handled through the AceDataCloud API rather than sent directly with an independently configured OpenAI API key.

Hosted connections support two authentication patterns. OAuth lets a compatible client sign in through a browser and approve access. A bearer token can be supplied when OAuth is unavailable or when an explicit integration credential is required. The documentation warns against configuring both methods for the same server. The hosted service may be metered, so pricing and account access should be checked before making real requests.

Setup and configuration

For hosted OAuth, add the MCP URL to a supported client and complete AceDataCloud sign-in. The README gives setup paths for Claude Desktop, Claude Code, Cursor, VS Code, Codex, Cline, and JetBrains AI Assistant. Some clients use an HTTP or remote-server setting, while Cline uses the streamableHttp transport value.

For bearer-token access, create an API key in the AceDataCloud application console and send it as Authorization: Bearer <token>. Keep the value in a client credential store or environment variable, not in prompts, screenshots, source control, or shared configuration. The local stdio route requires the mcp-openai-pro package and the ACEDATACLOUD_API_TOKEN environment variable.

The package name matters: the README identifies mcp-openai-pro as the AceDataCloud distribution and distinguishes it from an unrelated PyPI package named mcp-openai.

Tools and capabilities

The mcp-openai MCP server exposes these tool groups:

• Create chat completions and Responses API responses.
• Generate new images and edit existing images with AI.
• Produce text embedding vectors.
• Convert text to speech and transcribe audio from a URL.
• List chat, image, and embedding models.
• Retrieve the service usage guide.

These tools let an agent combine text generation with media generation, vector creation, and audio processing from one MCP connection.

Limitations and notes

Access to models and operations depends on the AceDataCloud application, enabled services, key restrictions, usage limits, account balance, and current pricing. A task lookup permission does not necessarily authorize generation requests. OAuth requires an AceDataCloud account, and dynamic client registration registers the client application rather than functioning as an API key.

The local process still calls the AceDataCloud API, so running it locally does not remove the service dependency. The README also notes that Claude Desktop's local configuration is separate from its hosted connector setup, and that client-specific environment-variable syntax is not interchangeable across Cursor, VS Code, and other clients.