BananaBanana

A hosted remote MCP server for AI image, video, and speech generation across models from Google, OpenAI, Alibaba, and xAI.

От сообществаРаботаетБез входаГлобальныйБесплатноМожет изменять данные

Что умеет

  • List Models: List all available image, video and speech generation models with current per-unit USD prices, supported resolutions, durations and constraints. Prices come from the same source as the we
  • Get Account: Get the current account balance (USD), this API key's name, optional daily spend cap and how much of it is used today. Free, no charge. Use it to check affordability before starting expen
  • Top Up: Get a balance top-up link. OAuth connections receive a one-time deposit-only link valid for 30 minutes; opening it cannot expose API keys, profile data or generation history. Bearer API-key us

Какие данные видит

Нужен ли аккаунт

Не нужен: сервер работает без входа

A hosted remote MCP server for AI image, video, and speech generation across models from Google, OpenAI, Alibaba, and xAI. Generate and edit images with Nano Banana 2 / Pro, GPT Image 2.5, and Qwen Image 3.0 Pro; create and edit video with Veo 3.1, Gemini Omni Flash, Wan 3.0, and Grok Imagine Video 1.5; and generate speech with Gemini TTS.

There is nothing to deploy and no provider API keys to configure — connect to the remote MCP endpoint with OAuth 2.1 or a BananaBanana Bearer API key from Claude, Cursor, VS Code, Windsurf, Codex, or any MCP-compatible client.

The server exposes 10 tools for model discovery, account and balance access, image generation and editing, video generation and editing, speech generation, job tracking, generation history, and balance top-ups. list_models returns the live model catalogue, capabilities, and current USD pricing so agents can choose models dynamically instead of relying on hard-coded prices.

Billing is pay-as-you-go with no subscription, seats, or expiring credits. Images start at $0.03, video at $0.09, and speech at $0.01 per started 200 transcript characters. Expensive video and multi-image operations return a cost quote before charging and require explicit confirmation, while failed or content-filtered generations are refunded automatically.

The same balance is shared between MCP and the BananaBanana web app. Top-ups are available with crypto from $1 or by card, PayPal, and SEPA from $20. For accountless machine-to-machine workflows, BananaBanana also provides an x402 endpoint for pay-per-call generation with USDC on Base.

Как подключить

Как подключить BananaBanana к Claude

Адрес сервера, который нужно вставить в настройках Claude:

https://bananabanana--support-dbuq.run.tools

  1. Откройте Claude (сайт claude.ai или приложение для компьютера).
  2. Перейдите в Настройки → Коннекторы (в английском интерфейсе: Settings → Connectors).
  3. Нажмите Добавить пользовательский коннектор (Add custom connector).
  4. Вставьте адрес сервера, скопированный ниже, в поле URL удалённого MCP-сервера и нажмите Добавить.
  5. В чате нажмите + → Коннекторы и включите новый коннектор.

Пользовательские коннекторы доступны на тарифах Free, Pro, Max, Team и Enterprise. На Free можно добавить только один. В Team и Enterprise коннектор сначала добавляет владелец организации в Настройки организации → Коннекторы.

Авторизация

После нажатия Добавить откроется окно входа в сервис. Войдите под своим аккаунтом и подтвердите доступ. Пароль Claude не видит.

Как подключить BananaBanana к ChatGPT

Адрес сервера, который нужно вставить в настройках ChatGPT:

https://bananabanana--support-dbuq.run.tools

  1. Откройте ChatGPT в браузере (chatgpt.com). Нужен тариф Plus, Pro, Business, Enterprise или Edu.
  2. Один раз включите режим разработчика: Настройки → Приложения → Дополнительные настройки → Режим разработчика (Settings → Apps → Advanced settings → Developer mode).
  3. Откройте Настройки → Коннекторы и нажмите Создать (Create).
  4. Заполните форму: Название (любое), Описание (одна фраза, что делает сервис), URL MCP-сервера (скопируйте ниже).
  5. В поле Аутентификация выберите OAuth, если сервис требует вход, иначе Нет. Нажмите Создать.
  6. В новом чате откройте + → Приложения/Коннекторы и включите коннектор.

Название раздела у OpenAI менялось (Connectors → Apps/Plugins). Если пункт называется иначе, ищите «режим разработчика» в настройках. В рабочих пространствах Business/Enterprise администратор должен разрешить пользовательские коннекторы.

Авторизация

При первом использовании ChatGPT откроет окно входа в сервис. Войдите и подтвердите доступ.

Как подключить BananaBanana к Cursor

Адрес сервера, который нужно вставить в настройках Cursor:

https://bananabanana--support-dbuq.run.tools

Быстрый способ: нажмите кнопку Открыть в Cursor ниже, Cursor сам предложит добавить сервер.

Вручную:

  1. В Cursor откройте Settings → Cursor Settings → MCP и нажмите Add new global MCP server.
  2. Вставьте JSON, скопированный ниже, в открывшийся файл ~/.cursor/mcp.json (для одного проекта: .cursor/mcp.json в корне проекта).
  3. Сохраните файл. Сервер появится в списке MCP; при необходимости пройдите авторизацию по кнопке рядом с ним.

Авторизация

Если сервис требует вход, рядом с сервером в списке MCP появится кнопка авторизации.

Список инструментов сервера (10)

Технические названия из tools/list. Нужны только разработчикам.

list_modelsList all available image, video and speech generation models with current per-unit USD prices, supported resolutions, durations and constraints. Prices come from the same source as the website — call this before quoting costs to a user or choosing a model. Free, no charge.
get_accountGet the current account balance (USD), this API key's name, optional daily spend cap and how much of it is used today. Free, no charge. Use it to check affordability before starting expensive generations. When the balance no longer covers even the cheapest image, the response also carries a ready-to-open top_up_url.
top_upGet a balance top-up link. OAuth connections receive a one-time deposit-only link valid for 30 minutes; opening it cannot expose API keys, profile data or generation history. Bearer API-key users receive the normal profile link. Free, no charge.
generate_imageStart an AI image generation (Google Nano Banana family, OpenAI GPT Image 2.5, Alibaba Qwen-Image 3.0 Pro). Charges the account balance immediately and returns a job_id — poll get_result for the finished image URLs. Typical completion: 10–60 seconds. The default nano-banana-2-lite model is the cheapest Google option and produces 1024px images; choose nano-banana-2 explicitly for 512px, 2048px or 4096px output. Optional reference_images provide the model with the actual subject, product, character or style pixels; Nano Banana Pro supports up to 14 references (per-model limits in list_models). Costs $0.03–$0.20 per image depending on model and resolution (see list_models). Failed generations are automatically refunded. relaxed_filter is true by default for Google images; pass false to opt out. Non-configurable checks still apply. Generating several images at once (number_of_images > 1) is a batch: the first call returns a price quote and charges nothing — repeat the call with confirm_cost set to the quoted amount to start. Example: {"prompt": "studio photo of a ceramic mug on linen, soft daylight", "model": "nano-banana-2-lite", "aspect_ratio": "4:5", "resolution": "1024"}
edit_imageEdit / refine a previously generated image with a text instruction on Nano Banana 2 Lite / 2 / Pro (multi-turn editing: change colors, remove objects, restyle, etc.). This tool modifies an image that already exists — use generate_image to create a new image from a prompt. Pass the job_id of a COMPLETED image generation as source_generation_id. Charged like a single image of the chosen model/resolution; auto-refund on failure. Example: {"source_generation_id": "cmxyz...", "prompt": "make the background pure white and add soft shadow"}
generate_videoStart an AI video generation (Google Veo 3.1 family, Gemini Omni Flash, Alibaba Wan 3.0 or xAI Grok Imagine Video 1.5). EXPENSIVE: $0.10–$6.00 per clip. Cost confirmation is mandatory: the first call always returns a USD quote and charges nothing — repeat the call with confirm_cost set to the quoted amount to actually start. Returns a job_id; poll get_result (videos take 1–10+ minutes). Failed generations are auto-refunded. Models: veo-3.1-fast (default, good quality/price), veo-3.1 (best Veo quality), veo-3.1-lite (cheapest, 720p/1080p), wan-3.0 (Alibaba Wan 3.0 — clips up to 30 s, 480p/720p/1080p, optional audio, first and last frame, seed, and VIDEO INPUTS via reference_videos: edit a clip, continue its story or carry its characters into a new scene, all driven by the prompt), grok-imagine-video-1.5 (xAI Grok Imagine Video 1.5 — 4–15 s at any whole second, 720p/1080p, native sound always on, priced per second by resolution — $0.14/s at 720p, $0.25/s at 1080p, input images free — aspect ratios 16:9 / 9:16 / 1:1 / 3:2 / 2:3, a first_frame at both 720p and 1080p, and up to 7 reference_images at 720p only — 1080p takes a first_frame but no references; no seed, last frame or video input), omni-flash (Gemini Omni 1.1 Flash — always has sound, 3–10 s, priced per second by resolution — $0.03/s at 360p up to $0.30/s at 4k; supports a last frame for interpolation, conversational editing via edit_from_generation_id, and scene extension via edit_video mode=extend), omni-flash-1.0 (the previous Omni generation, kept as an option: 720p at $0.10/s or 1080p upscaled at $0.12/s, 4/6/8/10 s, conversational editing, no last frame / scene extension / video references). Content filtering differs by model: omni-flash is by far the strictest, so a clip rejected as SAFETY_FILTERED on omni-flash is often produced by veo-3.1-fast without changing a word. Video has no configurable safety settings on Google's side (relaxed_filter is accepted for Veo but only pins its default personGeneratio
edit_videoEdit an EXISTING video (video-to-video). Default model omni-flash (Gemini Omni Flash): restyle it, replace or add objects, relight the scene, change the mood — motion and composition of the source clip are preserved. Billed by output length at $0.10/s; cost confirmation is mandatory (first call returns the quote and charges nothing). With model "wan-3.0" (Alibaba Wan 3.0) the source is read as a reference (up to 15 s): mode "edit" rebuilds the scene as the prompt says (framing follows the source, you pick the output length and resolution), mode "extend" shoots the next scene with the same characters and setting — a story continuation, not a frame-exact extension. Wan bills the SOURCE SECONDS TOGETHER WITH THE OUTPUT (the provider charges for both) and source + output must fit in 30 s. Source: either source_generation_id (a completed video from this account — see list_generations) or video_url (public http(s) link, max 200 MB). The source is normalised to MP4 720p and, for mode "edit", the FIRST 10 SECONDS (model limit); output is 720p with sound, aspect ratio follows the source. OUTPUT LENGTH ALWAYS EQUALS SOURCE LENGTH (the model cannot stretch or shorten a clip), so duration only works downwards: it trims the source to the first N seconds. Omit duration to edit the whole clip — the quote tells you the resolved length. mode: "extend" instead appends a continuation to the end of the clip — duration is then the LENGTH OF THE APPENDED PART (3–10 s), the source plays first and the total is capped at 40 s. Extension reads the WHOLE source (up to 37 s), so extensions chain: extend the result again until 40 s. Extension is billed on the full resulting length. With mode "extend", reference_images can bring a subject or product into the continuation ("the character from the reference image enters the scene"). Returns a job_id; poll get_result. Failed edits are auto-refunded. Example: {"prompt": "make the whole scene look like a pencil sketch, keep the motion identical", "so
generate_speechGenerate natural speech with Gemini 3.1 Flash TTS Preview. This synchronous tool returns a hosted WAV URL directly (no get_result polling). Supports one voice or an exactly two-speaker dialogue, automatic language detection or a BCP-47 language_code, natural-language direction for accent/tone/pace, and inline performance tags such as [whispers], [laughs], [very slow] and [excited]. Price is $0.01 per started 200 transcript characters; the account is charged only after Google has returned valid audio. Example: {"text":"[cheerfully] Welcome to BananaBanana!","voice":"Kore","style":"Warm product announcement, medium pace."}
get_resultGet the status and result of a generation job started with generate_image / edit_image / generate_video / edit_video. Waits up to wait_seconds for completion before returning (long-poll). On success returns hosted media URLs (valid 24 h — call again for fresh links), cost_charged_usd and balance_remaining_usd, plus a small inline preview for images. Free, no charge. Poll roughly every 10–15 s for videos.
list_generationsList this account's recent generations (both MCP and website) — id, type, model, status, cost and prompt preview. Use it to find a job_id to re-download results or to pick a source for edit_image / edit_video / generate_video edit_from_generation_id. Free, no charge.