FOR AGENTS · MPP

An image editor for AI agents, with MPP payments

“Make the mug blue and leave the rest alone.” A human gives an agent one sentence. The agent needs a way to turn it into an edited image, not another tool for the human to learn.

By Joshua Anderson · Published October 8, 2026 · Updated October 10, 2026

One destination for the editing job

Best Image Editor brings a documented image-editing interface, payment support, and result recovery onto one domain. An AI agent—a program that carries out tasks for a person—can discover the interface and submit an authorized edit without operating the browser editor.

The API, or application programming interface, is the set of requests the agent sends to the service. We publish plain-text agent instructions and an OpenAPI description, a machine-readable map of those requests and responses. An integration can start from a defined contract rather than guessing at website controls.

That is our case for an agent-friendly image editor: one place to perform the edit, clear payment terms, and a way to retrieve an existing result when a connection fails. Your agent still needs an authorized payment method and the required image inputs.

MPP makes payment part of the request

MPP stands for Machine Payments Protocol. It lets a service request payment in a form compatible software can respond to. Best Image Editor supports MPP card payments through Stripe, at $0.50 USD per edit at publication.

The agent checks the free capabilities endpoint for current availability and price. When payment is required, the service returns a payment challenge: instructions describing the payment needed for that request. An MPP-compatible client supplies an authorized payment credential and submits the same edit.

This route does not require a browser credit cookie, a site account, a cryptocurrency wallet, or the human’s image-provider key. It does require payment approval. The human should set a spending limit before the agent starts.

Give the agent a precise editing brief

The MPP edit accepts an image and a text instruction. After payment, it automatically selects the edit area and then edits the image, just like the browser’s Run action. The $0.50 payment covers both steps; no browser credits or separate selection request are needed. Check that the capabilities endpoint reports automaticMask: true before using automatic selection.

An agent can optionally supply its own mask to skip automatic selection. A mask is a matching-size image that marks where to edit: white selects the area, and black marks what should be preserved. If automatic selection fails or finds no matching area, the service stops before editing and starts a refund.

1. Keep the human’s brief. Record the requested change and the details that must remain.
2. Prepare the inputs. Provide the source image and a focused instruction. Omit the mask for automatic selection.
3. Check availability and budget. Confirm the price and obtain permission before paying.
4. Submit and retain recovery access. Store the payment credential securely before sending the paid request.
5. Inspect and deliver. Compare the result with the brief, download it, and show it to the human.

Recover the result instead of paying again

A timeout is not proof that an edit failed. The agent can use the original payment credential to request the existing result through GET /api/mpp/result. That recovery request does not buy another edit.

MPP results are private to the credential and do not enter the public gallery. Completed outputs expire after 15 minutes, so download them promptly. Temporary delivery is not permanent storage or a promise about every third party’s retention practices.

Follow the human’s request, then prove it visually

A beautiful image is not enough if the human asked for a different change. Check the target’s color, shape, edges, and relationship to the scene. Then inspect details the human asked to keep. Generated output is not a guarantee of exact preservation.

Use this workflow for campaign concepts, product styling, and other targeted revisions. Keep sensitive inputs out of the browser’s public workflow, and use human review before publishing a consequential edit.

Report what went wrong

If a request fails unexpectedly, the documentation is ambiguous, or the agent misreads something, it can tell us through the free POST /api/feedback endpoint. The agent describes what happened and what would have prevented it, without secrets, payment credentials, or image data. The team reviews every report and uses it to make the next agent’s job simpler.