deAPI MCP Server

What is deAPI MCP Server?

deAPI is a decentralized AI inference API: one REST surface for image generation, video generation, video character replacement, upscaling, speech and music synthesis, transcription, OCR and embeddings. This server exposes that surface to any MCP client, so a model can call it directly.

Two things make it comfortable for agents. Every job tool has a matching *_price tool that returns the exact cost of the call before any credits are spent, and every long job is polled for you with adaptive backoff and progress reporting - one tool call returns a finished result, not a job id to babysit.

How to use deAPI MCP Server?

Get an API key at app.deapi.ai/dashboard/api-keys - new accounts start with $5 in free credits - then point your client at the hosted endpoint:

{
  "mcpServers": {
    "deapi": {
      "type": "streamable-http",
      "url": "https://mcp.deapi.ai/mcp",
      "headers": {
        "Authorization": "Bearer <your-deapi-api-key>"
      }
    }
  }
}

In Claude Desktop and claude.ai you can instead add it as a connector and sign in with OAuth 2.0 (PKCE), with no key in a config file. Either way authentication is handled at the connection level - no tool takes an API token argument.

To self-host: Python 3.10+, uv pip install -e ., then python -m src.server_remote (port 8000), or use the provided Dockerfile and docker-compose.yml.

Start with get_available_models - it returns the live catalogue with per-model limits, defaults and capability flags. Model names should never be hardcoded.

Tools

Images

• text_to_image - generate an image from a text prompt
• text_to_image_price - cost of a text-to-image job
• image_to_image - transform an existing image with a prompt
• image_to_image_price - cost of an image-to-image job
• image_to_text - OCR and image understanding
• image_to_text_price - cost of an image-to-text job
• image_upscale - raise image resolution
• image_upscale_price - cost of an image upscale
• image_remove_background - cut the subject out of an image
• image_remove_background_price - cost of a background removal

Video

• text_to_video - generate a video from a text prompt
• text_to_video_price - cost of a text-to-video job
• image_to_video - animate a still image
• image_to_video_price - cost of an image-to-video job
• audio_to_video - drive a video from an audio track
• audio_to_video_price - cost of an audio-to-video job
• video_replace - replace the character in an existing video
• video_replace_price - cost of a video replace job
• video_upscale - raise video resolution
• video_upscale_price - cost of a video upscale

Audio

• text_to_audio - text to speech
• text_to_audio_price - cost of a text-to-speech job
• text_to_music - generate music from a description
• text_to_music_price - cost of a music generation job

Transcription

• audio_transcription - transcribe an uploaded audio file
• audio_transcription_price - cost of an audio transcription
• audio_url_transcription - transcribe audio from a URL
• audio_url_transcription_price - cost of an audio URL transcription
• video_file_transcription - transcribe an uploaded video file
• video_file_transcription_price - cost of a video file transcription
• video_url_transcription - transcribe a video from a URL (YouTube, X, Twitch and more)
• video_url_transcription_price - cost of a video URL transcription

Text

• text_to_embedding - embeddings for semantic search and RAG
• text_to_embedding_price - cost of an embedding job
• prompt_booster - expand a short prompt into a richer one before generating
• prompt_booster_price - cost of a prompt boost

Account and jobs

• get_balance - current credit balance
• get_available_models - live model catalogue with limits, defaults and capability flags
• check_job_status - status, progress and result of an async job

Key features of deAPI MCP Server

• Full deAPI v2 coverage - every endpoint in the API is reachable as a tool
• A free *_price tool for every job tool, so an agent knows the cost before it spends
• Adaptive polling with exponential backoff and progress reporting to the MCP client
• Runtime model discovery, with per-model limits, defaults and capability flags
• OAuth 2.0 with PKCE, or a Bearer token - authentication at the connection level, never in tool arguments
• Full Pydantic schema validation on requests and responses
• Automatic retries with error recovery on transient failures

Use cases of deAPI MCP Server

• Transcribe a podcast, a meeting recording or a YouTube, X or Twitch video into text
• Generate marketing images from a brief, then upscale them and cut out the background
• Turn a still product photo into a short clip, or replace the character in an existing video
• Voice over a script, or generate a soundtrack from a one-line description
• Read text off a scan, a screenshot or a photo with OCR
• Build semantic search or RAG over your own content with embeddings
• Price a generation before running it, and watch the credit balance from inside the agent