Overview
What is deAPI MCP Server?
A production-ready MCP server for the deAPI REST API. It exposes 39 deAPI AI capabilities as MCP tools, enabling LLMs to perform audio transcription, image generation, OCR, video generation, text-to-speech, and more. It is built for developers integrating AI features into LLM-based applications.
How to use deAPI MCP Server?
Install Python 3.10+, clone the repository, and install dependencies with uv pip install -e . (or pip/conda). Run the server locally with python -m src.server_remote (default port 8000) or deploy remotely with Docker. Connect from Claude Desktop via the built-in OAuth connector (using your deAPI API token) or via a config file with a Bearer token. Authentication is handled at the connection level, not per tool call.
Key features of deAPI MCP Server
- Complete API coverage with 39 MCP tools
- Smart adaptive polling for async jobs with exponential backoff
- OAuth 2.0 authentication with PKCE
- Automatic retry logic with error recovery
- Real-time progress reporting to MCP clients
- Full Pydantic schema validation for type safety
Use cases of deAPI MCP Server
- Transcribe audio or video files to text (including URLs from YouTube, Twitter, Twitch)
- Generate images and videos from text prompts or animate static images
- Perform OCR, remove backgrounds, upscale images and videos
- Convert text to natural speech or generate music from descriptions
- Generate text embeddings for semantic search and enhance prompts for AI inference
FAQ from deAPI MCP Server
What authentication does the server use?
OAuth 2.0 Authorization Code flow with PKCE (for remote connections) or Bearer token via HTTP headers (for local development). Tools do not accept an API token parameter; all authentication is handled at the connection level.
What are the runtime requirements?
Python 3.10 or higher with uv, pip, or conda. For remote deployment, Docker is supported via the provided Dockerfile and docker-compose.yml.
Where does data live?
The server forwards requests to the deAPI API and does not store API tokens. Tokens are used only for the specific request and never persisted or logged.
What transport does the server use?
Streamable HTTP (MCP over HTTP). The server can run locally on localhost:8000 or be deployed to a remote server with MCP_HOST=0.0.0.0.
Are there any known limits?
Polling timeouts are 5 minutes for audio and image jobs, 15 minutes for video jobs. The server retries failed requests up to 3 times
