MCP Server

Describes the dryAPI MCP server, its available tools, and how to connect agents or IDE integrations to use dryAPI as a structured toolbox.

WARNING

The MCP server is currently in beta. You may encounter occasional issues or unexpected behavior as we continue to refine the service. We actively welcome feedback and bug reports — join the conversation on Discord.

In addition to the HTTP API, dryAPI exposes an MCP (Model Context Protocol) server. This allows AI assistants, IDE extensions, and MCP-aware tools to call dryAPI capabilities as structured tools instead of manually constructing HTTP requests.

What is MCP?

The Model Context Protocol (MCP) is an open standard that enables AI models and agents to interact with external tools and services in a structured way. Instead of crafting raw HTTP requests, your AI assistant discovers available tools and calls them using a standardized protocol.

With dryAPI's MCP server, Claude, ChatGPT, Cursor, and other MCP-compatible clients can:

• Generate images, videos, and audio directly within conversations
• Transcribe YouTube / X / Twitch / Kick / TikTok videos or audio files
• Remove backgrounds and upscale images
• Check pricing before running expensive operations
• Access all dryAPI capabilities through natural language

New tools and capabilities will be added as the MCP server evolves. Follow our Discord for updates and to share feature requests.

The full source code for the MCP server is available on GitHub: github.com/dryapi-ai/mcp-server-dryapi

Server Details

Available Tools

The dryAPI MCP server exposes the following tools:

Image Generation & Manipulation

Video Generation & Processing

Audio & Speech

Embeddings

Utility

How to Connect

Claude (claude.ai)

To use dryAPI MCP tools directly in Claude conversations:

Step 1. Get your API key from the dryAPI Dashboard

Step 2. Add the MCP connector in Claude settings:

• Go to Settings → Connectors
• Click Add Connector or Add MCP Server
• Select New OAuth Connector
• Enter the following credentials:

Step 3. Grant permissions — After adding the connector, Claude will ask you to authorize the connection. Click Allow to enable dryAPI MCP tools in your conversations.

Step 4. Start using dryAPI MCP by asking Claude to generate images, transcribe videos, or use any other capability. Make sure to mention "dryAPI MCP" in your prompt so Claude knows to use the MCP tools.

Example prompts:

• "Using dryAPI MCP, generate an image of a futuristic city at sunset"
• "With dryAPI MCP, transcribe this YouTube video: [URL]"
• "With dryAPI MCP, transcribe this Twitch video: [URL]"
• "Check my dryAPI MCP balance"
• "Using dryAPI MCP, create a video of waves crashing on a beach"

Claude Desktop

To use dryAPI MCP in the Claude Desktop app, configure it via the config file with a Bearer token. This is the recommended approach for desktop — no OAuth setup required.

Config file setup:

Edit your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "dryapi": {
      "url": "https://mcp.dryapi.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Replace YOUR_API_KEY with your actual dryAPI key. Save the file and restart Claude Desktop.

Cursor IDE

To integrate dryAPI MCP into Cursor for AI-assisted development:

Step 1. Get your API key from the dryAPI Dashboard

Step 2. Open Cursor Settings:

• Press Cmd/Ctrl + Shift + P
• Search for "Cursor Settings" or navigate to Settings → Cursor Settings

Step 3. Add MCP Configuration:

• Find the MCP or Model Context Protocol section
• Add a custom MCP server configuration
• Paste the following JSON (replace YOUR_API_KEY with your actual key):

{
  "mcpServers": {
    "dryapi": {
      "name": "DRYAPI MCP",
      "url": "https://mcp.dryapi.dev/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Step 4. Restart Cursor to apply the configuration.

Step 5. Invoke dryAPI MCP in chat by using phrases like:

• "Using dryAPI MCP, generate an image of…"
• "With dryAPI MCP, transcribe this audio…"
• "Via dryAPI MCP, check my balance"

TIP

Cursor recognizes dryAPI MCP tools when you mention keywords like "using dryAPI MCP", "with dryAPI MCP", or "via MCP" in your prompts.

ChatGPT

To use dryAPI MCP tools in ChatGPT conversations:

INFO

MCP is available in beta to Pro, Plus, Business, Enterprise, and Education accounts. The MCP apps work on both web and desktop, but Developer mode settings are only accessible in the web version at chatgpt.com.

Step 1. Get your API key from the dryAPI Dashboard

Step 2. Enable Developer mode:

• Go to Settings → Apps & Connectors
• Scroll to Advanced settings at the bottom
• Enable Developer mode
• Go back — a Create App button will appear next to Advanced settings

Step 3. Create a new MCP App:

• Click Create App
• Fill in the following details:

Step 4. Check the risk acknowledgment checkbox ("I understand and want to continue")

Step 5. Click Create

Step 6. Start using dryAPI MCP by asking ChatGPT to generate images, transcribe videos, or use any other capability.

Example prompts:

• "Using dryAPI MCP, generate an image of a cyberpunk cat"
• "With dryAPI MCP, transcribe this YouTube video: [URL]"
• "Check my dryAPI MCP balance"
• "With dryAPI MCP, transcribe this Kick video: [URL]"

Self-Hosting

You can run the MCP server on your own infrastructure. The full source code is available at github.com/dryapi-ai/mcp-server-dryapi under the MIT license.

Prerequisites

• Python 3.10 or higher
• uv, pip, or conda for package management
• A dryAPI API key from the Dashboard

Quick Start

git clone https://github.com/dryapi-ai/mcp-server-dryapi.git
cd mcp-server-dryapi

# Install dependencies (uv recommended)
uv pip install -e .

# Start the server
python -m src.server_remote

The server will start on http://localhost:8000 by default. For external access:

MCP_HOST=0.0.0.0 MCP_PORT=8000 python -m src.server_remote

Docker Deployment

# Build and run
docker build -t mcp-server-dryapi .
docker run -d -p 8000:8000 --name mcp-server-dryapi mcp-server-dryapi

# Or with Docker Compose
docker-compose up -d

Cloud Deployment Options

The server can be deployed to any platform that supports Python or Docker containers: Railway, Fly.io, Heroku, DigitalOcean App Platform, AWS ECS, GCP Cloud Run, and others. Detailed deployment instructions are available in DEPLOYMENT.md in the repository.

Connecting a Self-Hosted Server

Point your MCP client to your server URL instead of the default https://mcp.dryapi.dev/mcp:

{
  "mcpServers": {
    "dryapi": {
      "url": "http://your-server-ip:8000/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Supported Models

The MCP server provides access to all dryAPI models.

Use the get_available_models tool to retrieve the full list with current specifications and limits.

Usage Examples

Generate an Image

Ask your AI assistant:

"Using dryAPI MCP, generate an image of an astronaut riding a horse on Mars, photorealistic style"

The assistant will call text_to_image with appropriate parameters and return the generated image URL.

Transcribe a YouTube Video

"With dryAPI MCP, transcribe this YouTube video with timestamps: https://www.youtube.com/watch?v=7qaiNfoMRts"

The assistant calls video_url_transcription and returns the full transcript.

Convert Text to Speech

"Using dryAPI MCP, convert this text to speech with a British female voice: 'Welcome to our platform'"

The assistant calls text_to_audio with the Kokoro model and appropriate voice settings.

Remove Background from Image

"Using dryAPI MCP, remove the background from this product photo"

The assistant calls image_remove_background and returns a transparent PNG.

Transcribe X/Twitter Spaces

"With dryAPI MCP, transcribe this Twitter Space: https://twitter.com/i/spaces/…"

The assistant calls audio_url_transcription and returns the full transcript.

Architecture

Smart Adaptive Polling

For async jobs, the server automatically polls for results with optimized intervals based on job type:

Polling uses exponential backoff (1.5× multiplier by default), so jobs receive fast initial checks while heavier workloads avoid unnecessary load.

Token Security

The MCP server does not store your API token. Authentication works at the connection level:

• Remote hosted server: OAuth 2.0 Authorization Code with PKCE, or Bearer token in HTTP headers
• Self-hosted server: Bearer token forwarded per-request to the dryAPI API
• Tokens are never persisted or logged; tools do not accept dryapi_api_token as a parameter

When to Use MCP vs HTTP API

Use the MCP server when you want an AI agent to decide which tools to call, with which parameters, and in what order. Use the HTTP API + queue for typical backend integrations where you directly manage requests and results.

Troubleshooting

Connection Issues

• Verify your API key is valid and has sufficient balance
• Check the server URL is exactly https://mcp.dryapi.dev/mcp
• For Claude: Ensure OAuth Client ID is dryapi-mcp and Secret is your API key
• For Cursor / Claude Desktop: Ensure Bearer token format — the Authorization header should be Bearer YOUR_API_KEY

Tools Not Appearing

• Restart your MCP client (Claude, ChatGPT, Cursor) after configuration changes
• In Claude: Make sure you clicked Allow when prompted to grant permissions
• In ChatGPT: Ensure Developer mode is enabled in Advanced settings
• Verify JSON syntax in configuration files (Cursor, Claude Desktop)
• Check that your API key has the necessary permissions

Jobs Stuck in Processing

• Use check_job_status tool with the job ID to monitor progress
• Large video generation or transcription jobs may take several minutes
• Check status.dryapi.dev for system status

Remote Connection Issues (Self-Hosted)

• Test the endpoint: curl -N http://your-server:8000/mcp
• Check server logs: docker logs mcp-server-dryapi or journalctl -u mcp-server-dryapi
• Verify firewall rules allow the configured port
• For SSL issues, confirm certificates are valid and paths are correct in your proxy config