X Image Splitter logoX Image Splitter
X Image Splitter logoX Image Splitter

Developers

Model Context Protocol (MCP)

Connect AI agents and client applications to %APP_NAME% through the Model Context Protocol.

X Image Splitter provides a remote Model Context Protocol (MCP) server over Streamable HTTP, exposing identical capabilities to the REST API. Connected AI Agents can check Credit balances, estimate and start Seedream 5.5 generations, improve prompts, find Creations, hand images to the Image Splitter and submit Support Requests.

Connection endpoint

Configure your MCP client with the server endpoint:

https://ximagesplitter.com/mcp

Connecting an account triggers OAuth authorization. Review the requested Scopes and confirm authorization. Once authorized, the client is registered under Account › Connected Apps, where access can be revoked at any time.

Public read operations (such as models and pricing) are accessible without authentication.

Authentication and security

  • Protocol: MCP endpoints enforce OAuth 2.1 authentication exclusively. API Keys are rejected on /mcp.
  • Resource indicator: Access tokens must target the https://ximagesplitter.com/mcp resource. Metadata is published at https://ximagesplitter.com/.well-known/oauth-protected-resource/mcp.
  • Scope validation: Invoking an action without the required Scope returns an insufficient_scope error indicating the missing Scope.
  • Revocation: Removing a Connected App in account settings immediately revokes token validity.

Available tools

The server publishes standard tool definitions via tools/list:

DomainTools
Productget_product_profile, list_available_models, get_model_capabilities, estimate_generation_cost, get_pricing
Accountget_account_profile, update_account_profile
Creditsget_credit_balance, get_credit_history
Generationcreate_generation, get_generation_status, list_generation_jobs, improve_prompt
Creationslist_creations, get_creation, update_creation_sharing, delete_creation
Uploadscreate_upload_target, get_upload_status
Billinglist_orders, get_order, create_checkout, get_subscription, change_subscription, cancel_subscription
Supportcreate_support_request, submit_refund_request, get_refund_status, submit_feedback

Each tool mirrors the corresponding REST API Integration Action. Webhook administration is not exposed via MCP; use Account › Webhooks or the REST API.

Execution model and human hand-offs

  • Cost estimation: Clients should invoke estimate_generation_cost prior to create_generation. Generation Jobs execute asynchronously; clients poll get_generation_status or listen for webhook notifications.
  • Prompt improvement: improve_prompt costs 1 Credit, charged only when a suggestion is returned. Tell the user the cost and get their agreement before calling it, and send an idempotency key so a retry is charged once.
  • Image Splitter hand-off: get_generation_status and get_creation return splitterLinks, one per finished image. Give the user the link; they open it signed in to the same account and the image is split in their browser. Splitting never runs on the server, and there is no splitting tool.
  • Financial operations: create_checkout returns a URL where the user picks a payment method and pays for a Credit Pack themselves. Subscriptions are no longer sold; get_subscription, change_subscription and cancel_subscription exist only for subscriptions created earlier, and also return confirmation URLs. Agents cannot complete a payment.
  • Destructive operations: delete_creation permanently hides a Creation and revokes its Share Link.

Example prompts

How many credits do I have left, and when do they expire?
Make a 16:9 panorama of a lighthouse coast at sunset in X Carousel Ready mode.
Tell me the price first, then give me the link to split it.
Improve my prompt "five lipstick tubes in a row" if it costs 1 Credit.

Protocol specifications

The server implements MCP protocol revision 2026-07-28 over Streamable HTTP (POST only, stateless). Clients register via Client ID Metadata Documents as described in Connected Apps & OAuth. Every response includes an x-request-id header for diagnostic tracing.