MCP App Store
by KeenEthics
Batch Banana icon

Batch Banana MCP Server for ChatGPT

By Lowside Labs, Inc.

Overview

Batch Banana generates images and video from text prompts and reference images, and edits images you already have. What you can do: - Generate images with a choice of image models, aspect ratios, and up to 8 images per request. - Generate short videos from a prompt, a start frame, or reference images. - Edit images: remove backgrounds, upscale, relight, expand, restyle, and make text-guided edits. - Search your Batch Banana library to reuse earlier images as references, and file new work under your projects. How it works: the assistant lists the available models, starts the job, and returns links to the finished images or video. Every output is also saved to your Batch Banana library. Requirements: a Batch Banana account. You sign in on batchbanana.com and choose which workspace to connect. Limitations: generation runs at third-party model providers, which refuse requests that break their content policies. Videos can take several minutes to finish.

How to connect

  1. ChatGPT: Open the Batch Banana listing in the ChatGPT app directory and choose Connect. ChatGPT listing Batch Banana exposes 33 tools here.

Tools

cancel_batch

ChatGPT
Stop a batch's running run, only when the user asks to. Images already started finish, rows not started are skipped, and their credits are refunded. A batch with nothing running is left as it is.

complete_upload

ChatGPT
Finish an upload after its file was PUT to uploadUrl. A reference image comes back as a file whose id goes in referenceFileIds; a batch CSV's upload id goes in a batch tool's csvUploadId.

create_batch

ChatGPT
Create a batch: a table of prompts that run together, one or more images per row, up to 20000 rows. Rows come in two ways. Send up to 500 rows directly in rows. For more, if you can send an HTTP PUT, write a CSV whose header line is prompt,model,aspect_ratio,images,style,references,options (only prompt is required), upload it with create_upload (purpose batch_csv), PUT it and call complete_upload, then pass its id as csvUploadId. A CSV imports in the background: get_batch with waitSeconds until csvImport.status is completed, or failed with every problem by line. Pass run.expectedCreditCost (totalCredits from quote_batch) to start running every row at once, or once a CSV's rows are written; the run starts preparing, is charged once when every row is priced, and is refused with nothing charged if the price moved. Requires idempotencyKey. Order: quote_batch, then create_batch with run.expectedCreditCost set to the quote's totalCredits (or update_batch with action run), then get_batch with waitSeconds until the run ends, then read results, then update_batch with action retry_failed for rows that failed.

create_project

ChatGPT
Create a project in the workspace and return its id. Pass the id as projectId on a generate, edit or batch tool to file the work under it. Requires idempotencyKey; a retry with the same key returns the same project.

create_upload

ChatGPT
Start an upload for a file too large to send inline: an image (purpose reference) or a batch's rows as CSV or TSV up to 25 MB (purpose batch_csv). Returns an id and an uploadUrl. PUT the file's bytes to uploadUrl with the same Content-Type, then call complete_upload with the id. Needs a client that can send an HTTP PUT, such as a shell with curl; otherwise use upload_image, or send batch rows directly.

describe_edit_operation

ChatGPT
Full parameter definitions for one edit operation. Use the id from list_edit_operations.

describe_model

ChatGPT
Full parameter definitions for one model: options, resolution choices, and capabilities. Use the id from list_models. Every parameter is sent under options on quote and create, keyed by its key, including resolution, as options.resolution.

download_files

ChatGPT
Get download links for a batch's images as ZIP files of up to 1,000 rows each, every file named by its row index and column. The links work for an hour; give them to the user to open.

edit_image

ChatGPT
Run an edit operation on stored images: fileIds, up to 256, or up to 40 for an operation that costs credits, which runs at an AI provider. Requires idempotencyKey. An operation's sourceArity (list_edit_operations) says how the images become jobs: each makes one job per image, the result lists every id in jobIds, and each is polled with get_generation; one and all make one job, which may hold many outputs. then runs one follow-up operation on every output (for example remove the background, then make stickers). It starts after this request settles: poll get_generation on a job until followUp appears, then poll each id in followUp.jobIds for the follow-up's outputs. A request whose jobs all failed gets no followUp, so stop polling once every job has failed. Quote first with quote_image_edit: many images cost the per-image price times the count.

generate_image

ChatGPT
Create an image generation. prompt takes one prompt, or a list of distinct prompts for a set: concepts, directions, or the items of a series. Send a set as one call with every prompt, never one call per prompt, so the user sees it together as one result. numImages is per prompt. Without it, one prompt, alone or as a one-item list, makes 4 images, and each of several prompts makes 1. Requires idempotencyKey (reusing it dedupes an identical request; a different request under the same key is a conflict). Credits are charged up front and refunded on failure. style takes a slug from search_styles. Returns the first job, and every job in jobIds, in prompt order, when there are several; poll get_generation for output URLs.

generate_video

ChatGPT
Create a video generation. Requires idempotencyKey. Returns a job; poll get_generation for the output URL.

get_batch

ChatGPT
Check a batch: one line of progress, and optionally a page of results with output URLs (results.status failed lists what failed and why). waitSeconds waits for the run or a CSV import to move, so call it in a loop rather than polling fast. The batch has settled when inProgress is false; a CSV import that failed lists its problems by line, with the fix for each. Where the Batch Banana results view shows the batch, the user already sees its progress and images, so describe them rather than embed their URLs. Order: quote_batch, then create_batch with run.expectedCreditCost set to the quote's totalCredits (or update_batch with action run), then get_batch with waitSeconds until the run ends, then read results, then update_batch with action retry_failed for rows that failed.

get_file

ChatGPT
Fetch one stored file's URLs, content type, size, and dimensions by file id.

get_generation

ChatGPT
Fetch a job's current status and outputs (URLs). Poll this after a generate or edit tool until status is completed or failed. This is the polling tool; there is no push subscription. Where the Batch Banana results view shows the job, the user already sees each output there as it finishes, so describe the result rather than embed its URLs.

get_profile

ChatGPT
Return the account and workspace this connection acts as. The opaque id stays the same across token refresh and reconnection.

import_image

ChatGPT
Import a public https image URL into the workspace and return a file id. Generation and edit tools take file ids, not URLs, so import first. Reusing idempotencyKey returns the original file.

list_edit_operations

ChatGPT
List image edit operations (background removal, upscales, relight, expand, and more) with requirements and cost.

list_models

ChatGPT
List image and video generation models with pricing and capabilities. Optionally narrow to one media type.

list_project_jobs

ChatGPT
List the jobs filed under one project, newest first, with their status and outputs. Pass nextCursor back as cursor for the next page.

list_projects

ChatGPT
List the workspace's projects, newest first. Pass a project id as projectId on a generate or edit tool to file the job under it, or on search_library to search only that project.

list_reference_sets

ChatGPT
List the user's saved reference sets: named groups of images such as a project's style, a character or a product, each with its type, its note on how to use it, and its images. With projectId, lists that project's sets and then the workspace's; without it, only the workspace's. Call it when the user names a set or asks to keep a look, character or product consistent, then pass the set's id or exact name in referenceSets on generate_image, generate_video or a batch filed under the same project. A set marked Auto-attach already goes on every image request in its project, so leave it out of referenceSets. The note is sent with the prompt, so the prompt need not repeat it.

list_workspaces

ChatGPT
List the user's Batch Banana workspaces this connection can use, each with its id, plan and credits, and which one is current. Every tool acts in the current workspace: its credits pay, and new work lands in its library.

quote_batch

ChatGPT
Price a batch without charging or writing anything, and list rows that can't run with the field and fix for each. Send defaults with rows or a csvUploadId for a batch you have not made yet, or the id of a batch to price a run of its rows (rowIndexes for some, retry failed for a retry). A row takes a prompt and may set its own model, aspectRatio, numImages, style (a slug from search_styles) and references (file ids or https image links, imported only when the rows are written). Rows come in two ways. Send up to 500 rows directly in rows. For more, if you can send an HTTP PUT, write a CSV whose header line is prompt,model,aspect_ratio,images,style,references,options (only prompt is required), upload it with create_upload (purpose batch_csv), PUT it and call complete_upload, then pass its id as csvUploadId. A CSV imports in the background: get_batch with waitSeconds until csvImport.status is completed, or failed with every problem by line. Order: quote_batch, then create_batch with run.expectedCreditCost set to the quote's totalCredits (or update_batch with action run), then get_batch with waitSeconds until the run ends, then read results, then update_batch with action retry_failed for rows that failed.

quote_image_edit

ChatGPT
Price an image edit without charging. Returns creditCost, pricingVersion, and requestHash. Pass creditCost back as expectedCreditCost on the matching create tool to guard against a price change between the quote and the create.

quote_image_generation

ChatGPT
Price an image generation without charging. Returns creditCost, pricingVersion, and requestHash. Pass creditCost back as expectedCreditCost on the matching create tool to guard against a price change between the quote and the create.

quote_video_generation

ChatGPT
Price a video generation without charging. Returns creditCost, pricingVersion, and requestHash. Pass creditCost back as expectedCreditCost on the matching create tool to guard against a price change between the quote and the create.

read_results

ChatGPT
For the Batch Banana results view only: reads jobs by id with their output URLs, so the view can follow running jobs.

search_library

ChatGPT
Find the workspace's existing images and video, one row per job with its outputs, cost, model and references. Filters: kind, model, operation, createdAfter and createdBefore, projectId, mediaType. query searches prompts and orders by relevance; without it, jobs list newest first. Most edits (stickers, upscales) and every import have no prompt, so query misses them; filter by kind or operation instead. The date filters do not combine with query. model matches generations; an edit records its operation as its model, so an operation id in model matches those edits too. limit is about how many jobs a page holds, up to 100. A page can be short or empty while nextCursor is set, especially with model or operation filters; keep paging until nextCursor is null. An output's fileId works as a referenceFileIds entry, and jobId works with get_generation. Pass nextCursor back as cursor for the next page.

search_styles

ChatGPT
Search the Batch Banana style library: curated looks such as watercolor, claymation or 35mm film. query describes a look and returns the closest styles, a line each, the styles the query names first. Each match's match field is name when the query names that style, or look when the style is only the nearest by its images, so read a look match's look line to judge whether it fits. slugs reads up to 6 chosen styles in full: description, what they avoid, similar styles. Neither lists the whole library by family, names only. A note in the answer says how to read it. Any style goes by slug in generate_image's style (and the quote's), which adds its reference images and description on any model that takes reference images, so the prompt needs only the subject. When nothing fits, write the look into the prompt instead. These are Batch Banana's styles, not the workspace's own work: search_library finds that.

show_results

ChatGPT
Show the user generations and edits, by job id, as images and video in the Batch Banana results view. Use it whenever the user wants to see work again: images made earlier in the conversation, or jobs found with search_library or list_project_jobs. The view loads every output and follows jobs still running. The view is how the user sees them: the reply adds no Markdown images, links to them, other images, or image search results.

switch_workspace

ChatGPT
Make another of the user's workspaces the current one, by its id from list_workspaces. Every tool acts in the current workspace, so later work, files and credits go there until the user switches again. Switch only when the user asks you to or agrees to, never on your own to get around a refusal.

update_batch

ChatGPT
Change a batch with exactly one action: add_rows (up to 500 rows, or a CSV upload's csvUploadId), run (every row, or rowIndexes), or retry_failed (failed rows, in columnId or in every column). run and retry_failed take expectedCreditCost from quote_batch for the same rows and start preparing; only one run goes at a time. Requires idempotencyKey. To stop a run, use cancel_batch. Order: quote_batch, then create_batch with run.expectedCreditCost set to the quote's totalCredits (or update_batch with action run), then get_batch with waitSeconds until the run ends, then read results, then update_batch with action retry_failed for rows that failed, and download_files for ZIP links the user opens to save the images.

upload_image

ChatGPT
Upload an image you have as base64 (PNG, JPEG, WebP or GIF, up to 5 MB) and get a file id to pass in referenceFileIds. For a public URL use import_image. For a larger file, or a batch CSV, use create_upload and complete_upload if you can send an HTTP PUT.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/batch-banana"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/batch-banana)