A side-by-side look at the Nano Banana MCP and Distribea MCP MCP servers: transport, auth, tools, quality signals, and the config to install either one in Claude, Cursor, Windsurf or VS Code.
Verdict: choose Nano Banana MCP for multimedia process work as a local process; choose Distribea MCP for multimedia process as a local process. You can also run both side by side in one mcpServers block.
generate_image, edit_image, continue_editing.DISTRIBEA_MCP_KEY.make_images, bring_alive, generate_image.Rows where the two servers differ are marked.
| Spec | N Nano Banana MCP | |
|---|---|---|
| Category | Multimedia Process | Multimedia Process |
| Quality signal | 55/100 · Good | 57/100 · Good |
| Install command | Confirmed | Confirmed |
| Transport | Local subprocess (stdio) | Local subprocess (stdio) |
| Auth | Not declared | Not declared |
| Pricing | Not declared | Paid service |
| Required env vars | None listed | DISTRIBEA_MCP_KEY |
| Verified owner | Community listing | Community listing |
| GitHub stars | 0 | 1 |
| Directory activity | 0 views 0 installs 0 upvotes | 1 views 0 installs 0 upvotes |
| Tools | 6 | 21 |
| Full listing | View Nano Banana MCP | View Distribea MCP |
generate_imageCreate new images from text promptsedit_imageModify existing images with optional reference imagescontinue_editingIteratively edit the last imageget_last_image_infoGet info about the last imageconfigure_gemini_tokenSet API keyget_configuration_statusCheck configurationmake_images⭐ THE flagship one-call tool. page_path → dresses that page; no page_path → scans the WHOLE project. Finds every placeholder/stock slot (placehold.co, picsum, unsplash, pexels…), locks/infers the style automatically, generates every image IN PARALLEL (recurring characters/products auto-used), saves optimised WebP and patches src+alt in the code. REVIEW/TESTIMONIAL sections are detected automatically: their avatars come out as ultra-real casual smartphone selfies (UGC look — car/bedroom/living-room backdrop, real skin), each reviewer keeps the SAME face across THIS site; your other sites are independent by default (set cross_site_unique:true to also keep reviewer faces distinct across ALL your sites). rebrand:true targets EXISTING real images instead: first call lists them for FREE, then apply:true regenerates them all in place (code untouched, originals kept as *.original).bring_aliveMake an EXISTING site feel alive with on-brand imagery. Two things at once: (1) ADDS images where sections have none (hero, testimonials with ONE face per reviewer, team/about, key benefits, gallery — sized to the real layout), and (2) by DEFAULT REPLACES every existing image with an on-brand one (stock photos, old pictures…). page_path = one page; OMIT page_path = the WHOLE site. WORKFLOW: call it once for a FREE proposal (new spots + the list of existing images that will be replaced, 0 credits); show it, ask the user which to KEEP; then call AGAIN with apply:true to place + fill the new spots AND replace the existing images in one go. Spare specific images with keep:["filename"]; pass replace_existing:false to only ADD images and leave existing ones untouched. Everything is reversible with undo:true.generate_imageGenerate ONE website image for a SPECIFIC spot AND wire it straight into the page — DESTINATION = a page (placed in code), NOT a file on disk. Coherent with the locked style (and optionally a recurring character/product). Generation is tied to a placement: pass page_path + leave a placeholder (<img src="https://placehold.co/…">) where you want it; the image replaces that placeholder. NO page / NO placeholder = NOTHING is generated and NOTHING is billed (never produces an orphan image). Subjects mentioning a customer review/testimonial/avatar automatically switch to the UGC mode: ultra-real casual smartphone selfie of an everyday person (unique per site, same reviewer = same face). For a whole page or project, prefer make_images. CONTEXT: when you are building or editing a website (pages, components, placeholders), THIS or make_images is the default — that is the page-building job. If the user instead wants a downloadable image FILE to keep/reuse (an ad, a social/marketing visual, "download/save the png", a named model with no page), use generate_with_model. If it is unclear whether the image goes IN a page or is a downloaded FILE, ask one short question first.list_modelsReturns the catalog of image models to choose from: id, name, what it's best at, quality tiers, formats, max images per call, and an indicative price in credits. Call this BEFORE generate_with_model when the user wants to pick a specific model or compare options/prices. Image only for now.generate_with_modelStandalone image generator — DESTINATION = a FILE on the user's disk, never a website page. Use it for downloadable assets the user keeps or reuses elsewhere: ads, social posts, marketing/promo visuals, thumbnails, a reusable hero or logo, or any 'generate then download / save to a folder' request (often with a named model like Flux Pro / Seedream, or a png/jpg format). DESTINATION ALWAYS WINS: if the image must appear ON a website page — or you are MID-BUILD of a site, filling placeholders / dressing pages — use generate_image (one spot) or make_images (whole page), EVEN IF a model is named. If it is genuinely unclear whether the user wants the image IN a page or as a downloaded FILE, ASK one short question before generating — do NOT guess. Downloads a NATIVE png/jpg (never WebP) into a tidy folder, at the regular SITE price; nothing is placed into any page. Get model ids from list_models. DEFAULT when the user names NO model: nano-banana-pro at 2K. ORIENTATION: if the user didn't state one, ASK 'portrait or landscape?' before generating, then pass the matching format (16:9 = landscape, 9:16 = portrait). (Same idea as Higgsfield: pick a model, dial settings, download the file.)list_video_modelsReturns the catalog of VIDEO models to choose from: id, name, what it's best at, durations (seconds), resolutions, aspect ratios, and an indicative price in credits (for the default duration; longer clips cost more). Call this BEFORE generate_video when the user wants to pick a specific video model or compare options/prices.generate_videoStandalone VIDEO generator — DESTINATION = an mp4 FILE on the user's disk. Pick a specific video model + settings (duration, resolution, aspect ratio) and DOWNLOAD the result as a native mp4 into a tidy folder, billed at the regular SITE price. Video takes several MINUTES and needs a PAID plan (no free video). Use it for ads, social clips, b-roll, promos to download — never to place into a website page. If the call says the video is still rendering, run it again with the SAME settings in ~1 min to pick it up (no double charge). Get valid model ids + options from list_video_models. DEFAULT when the user names NO model: grok-imagine-video, 6s, 720p. ORIENTATION: if the user didn't state one, ASK 'vertical or horizontal?' before generating, then pass the matching format (16:9 = horizontal, 9:16 = vertical).list_music_modelsReturns the catalog of MUSIC models (id, name, indicative price in credits). Call this BEFORE generate_music when the user wants to pick a specific music model or compare prices.generate_musicStandalone MUSIC generator — DESTINATION = an audio FILE on the user's disk (mp3), downloaded into a tidy folder, billed at the SITE price. Needs a PAID plan (no free audio). Use it for background tracks, jingles, b-roll music to download — never placed into a page. Describe the music (style, mood, instruments) in `subject`. If it says still rendering, run again with the SAME settings to pick it up (no double charge). Get model ids from list_music_models. DEFAULT when the user names NO model: minimax-music.list_voice_modelsReturns the catalog of VOICE-OVER models: id, name, supported languages, number of voices, a few sample voice ids, and an indicative price per 500 characters. Call this BEFORE generate_voice to pick a model + a voice id + a language.generate_voiceStandalone VOICE-OVER generator — DESTINATION = an audio FILE on the user's disk (mp3), downloaded into a tidy folder, billed at the SITE price (per character). Needs a PAID plan (no free audio). Use it for narration, ads, IVR, video voice-overs to download — never placed into a page. Pass the script in `text`, plus a voice id and language from list_voice_models (a default voice is picked if you omit it). If it says still rendering, run again with the SAME settings to pick it up (no double charge). DEFAULT when the user names NO model: elevenlabs-multilingual-v2 with the Roger voice.blog_coverAUTONOMOUS TRIGGER — call this when an EDITORIAL ARTICLE still LACKS a cover/hero image (a brand-new post, or one left with a placeholder/stock image), even if the user never mentions an image. GUARDRAIL against noise: fire it ONCE per article, only when the cover is missing; if the article already has a real cover, leave it — do NOT regenerate; a pure text/typo/SEO edit is NOT a trigger; never re-run a successful call or 'improve' a cover on your own — only on the user's explicit request (each call bills the subscriber). Recognise the article by MEANING, not by tags: any standalone piece written to be READ — a headline + a body of prose on one topic (blog post, news, guide, tutorial, case study, magazine/journal piece) — in ANY language, on ANY stack (markdown/MDX, a headless CMS like WordPress/Sanity/Contentful, or a custom DB-driven route). Folders/routes (/blog, /posts, /articles, /news, /journal, /actualités, /magazine, /guides), markdown/MDX frontmatter, or an <article> element are only NON-EXHAUSTIVE hints — none is required: if it simply reads like an article, it qualifies (a marketing/landing/product page does NOT — that keeps make_images). You already hold the article you are working on → pass its own title + text as article_text (or article_url for a public link, or article_file for a local md/mdx/html/txt file). It generates the COVER, driven by the ARTICLE'S OWN specific subject (never a generic trade photo), kept visually coherent with the site's locked style (palette, light, medium), adapted to the site's country (driving side, architecture, currency), with NO recurring face/logo/product stamped on by default (every article gets a DIFFERENT scene — no 'same baker's face on every post'). Default = the cover ONLY at 16:9; set illustrations:N to also get N in-article images. Delivers an optimised WebP + ALT, ready to host. To DELIBERATELY feature a locked character/product, pass character/product. If the site has no locked style it infers one from the project; if it still lacks the trade/brand it tells you what to provide.Paste either block (or both) into your client config, e.g. claude_desktop_config.json or ~/.cursor/mcp.json. We only show a config when the install command is confirmed.
{
"mcpServers": {
"nano-banana-mcp-2": {
"command": "npx",
"args": [
"-y",
"github:fjacquet/Nano-Banana-MCP"
]
}
}
}
{
"mcpServers": {
"distribea-mcp": {
"command": "npx",
"args": [
"-y",
"distribea-mcp@latest"
],
"env": {
"DISTRIBEA_MCP_KEY": "YOUR_DISTRIBEA_MCP_KEY_HERE"
}
}
}
}