Skip to main content
Recraft supports the Model Context Protocol (MCP), enabling AI agents such as Claude, Cursor, and other MCP-compatible clients to interact with Recraft’s image generation and editing tools through a standardized interface. By connecting Recraft’s MCP server to your client, you can generate and edit high-quality raster and vector images directly from your agent environment without switching tools. Recraft offers to use a remote MCP server at https://mcp.recraft.ai/mcp that requires no local installation. Authenticate via OAuth with your Recraft account and connect directly from any MCP-compatible client. Usage is billed against your Recraft subscription credits (the same balance used in the Recraft Studio app). See MCP server for setup instructions.

​
Supported operations

Recraft’s MCP server expose the following operations:
  • Raster and vector image generation
  • Raster and vector image transforms
  • Creating and using custom styles
  • Vectorization of raster images
  • Background removal and replacement
  • Upscaling of raster images
See MCP server for setup details.

​
Supported clients

Any MCP-compatible client that supports the Streamable HTTP transport and OAuth can use the remote server.

​
Troubleshooting

  • OAuth window doesn’t open or loops back. Clear the cached OAuth state in your client and reconnect. For mcp-remote, delete ~/.mcp-auth/ (or the equivalent directory for your OS) and restart the client.
  • Tools don’t appear in the client. Confirm the connector is enabled in your client’s tools menu. In Claude Code, run /mcp to verify the server is connected and authenticated.
  • “Insufficient credits” errors. Top up your subscription credits on the Recraft web platform. MCP calls deduct from this balance, not from your API units.
  • Generation succeeds but no image preview is shown. Some clients impose a size limit on inline images. Retry with a smaller size, or open the returned URL directly.
  • Tools call appears to hang. Image operations can take up to a few minutes for large outputs or upscales. Wait for the call to time out before retrying.