MCP App Store
by KeenEthics
Synthesys icon

Synthesys MCP Server for ChatGPT

By Synthesys

Overview

Create finished videos from a short brief: product ads, UGC-style videos, explainers and avatar videos with voiceover, music and captions. Generate short video clips and images, recreate an existing ad for your brand from a link, and use your own uploads, avatars, voices and Brand Kit. You review and confirm each step before anything is created. Sign in with your Synthesys account to get started.

How to connect

  1. ChatGPT: Open the Synthesys listing in the ChatGPT app directory and choose Connect. ChatGPT listing Synthesys exposes 17 tools here.

Tools

answer_question

ChatGPT
When get_video_status shows state 'awaiting_answer', the build is paused on a clarification question (pending_question). Relay the question to the user, then send their reply here to resume the build — or pass skip:true to let the agent use its best judgment. Free and instant. Only for QUESTION gates; a pending cost approval is confirmed via create_video {project_id, confirm:true}. If the question changed before the answer landed, a typed answer is NOT forwarded to the new question — the result (state 'question_changed') carries the new question to relay instead.

create_video

ChatGPT
PAID, two-step. Build a COMPLETE multi-scene Synthesys video (script, scenes, voiceover, avatars, captions, music, final composite) from a brief. Step 1 — call with a brief: the agent plans the video (free) and this returns a COST QUOTE (scenes, seconds, credits; 100 credits = $1) with state 'quote_ready'. Nothing is spent yet. Planning usually takes 30-90 s and this call waits for it; if it returns state 'planning', keep polling get_video_status until the quote appears. Step 2 — ONLY after the user explicitly accepts the quote, call again with {project_id, confirm: true}: the FULL production then builds hands-free, usually taking 10-25 MINUTES. Poll get_video_status every 30-60 seconds and do not give up before 30 minutes. If the user says no, call {project_id, confirm: false} — that DECLINES the pending step and spends nothing. Calling with {project_id} alone (no brief, no confirm) is a safe read-only status check that approves nothing. ONLY confirm:true ever spends money. If status shows a pending question, relay it to the user and reply via answer_question. To put a specific person on camera, choose the narrator or make the video on-brand, pass actor_id (list_actors), voice_id (list_voices) or brand_kit_id (list_brand_kits) with the brief; leave them out and Synthesys chooses. Free settings the build takes straight from the brief (for example switching the narrator off for a music-only video) are applied for you and listed in applied_free_settings, so the first thing you are asked to confirm is the price.

export_video

ChatGPT
Free (0 credits), with a daily limit per account. Render the saved edits into a finished MP4. The first call starts it and returns a job_id; call again with project_id and job_id every 15-30 seconds until state is 'complete', then share video_url. Each export becomes a new version of the video and never replaces an older one. Plans decide what's allowed: free accounts export 720p with a watermark. Wait for get_edit_status to report 'complete' first, or the export won't include those edits.

generate_clip

ChatGPT
PAID. Render ONE short AI video clip (4-15 s) from a text prompt — the single-shot option (cinematic product shot, b-roll, a one-scene ad). Starts an autonomous render and returns a project_id handle IMMEDIATELY; the render itself typically takes 8-13 minutes. Poll get_video_status with the project_id every 30-60 seconds until state is 'complete', then share video_url. Cost: roughly 40-50 credits per second at the default cinematic quality (~$0.40-0.50/s, about 200 credits for 4 s); the response includes an approx_cost_credits estimate for the chosen template and length, and the finished status reports what was actually charged. The default template is product-only but may show hands; pass no_people:true when nothing human should appear. Attach the user's product shots/footage via reference_asset_urls (from upload_reference). For multi-scene productions with voiceover/captions/music, use create_video instead. TO CHOOSE THE MODEL: pass a model (a model_id from list_models {category: 'video'}). That path is QUOTE-FIRST instead: the first call returns the exact price for that model, length and resolution plus a quote_id, and spends nothing; only after the user says yes, call again with the same inputs plus confirm: true, that quote_id and quoted_credits (the price they saw). One quote starts at most ONE clip: if that call errors or times out, call again with the SAME quote_id (it returns the same clip, never a second charge) — never ask for a new quote to retry. It then returns a project_id to poll with get_video_status; the clip takes a few minutes and is charged only when delivered. With a model, use start_frame_url for a picture to animate; template and reference_asset_urls do not apply.

generate_image

ChatGPT
PAID, two-step. Generate one or more images from a prompt — or edit an image the user provides — with a chosen model. Step 1 — call with the prompt (and optionally model, aspect_ratio, resolution, images, reference_asset_urls): it returns the EXACT price in credits for those settings and spends nothing (state 'quote_ready', with a quote_id). Step 2 — ONLY after the user explicitly accepts the price, call again with the SAME inputs plus confirm: true, the quote_id and quoted_credits (the price they saw): the images are generated in seconds to a couple of minutes and returned as asset_url links. One quote pays for at most ONE press: if that call errors or times out, call again with the SAME quote_id — it returns the same press (waiting for it if it is still running) and never charges twice; never ask for a new quote to retry. The charge is exactly the quote; an image that fails costs nothing; if the settings changed the price, nothing is charged and the new price is returned to show the user. Every image is saved in the user's Synthesys library, so its asset_url can go straight into reference_asset_urls of create_video or generate_clip (for example: generate a product shot, then build the ad around it). Omit model to use the default; call list_models {category: 'image'} to choose one and see prices. To EDIT an image, pass it in reference_asset_urls with edit: true and a model whose can_edit_images is true.

get_credits

ChatGPT
Get the account's credit balance, the price schedule, recent usage and — when the account has one — the daily spending limit for assistant/API use (daily_cap, spent_today, remaining_today; it resets at 00:00 UTC). Free and instant (<2 s). 100 credits = $1. Check this before paid tools (generate_clip, create_video, recreate_from_url) so you can warn the user if the balance or today's remaining limit won't cover the cost. A balance of 'unlimited' means a test backend without credit enforcement.

get_video_status

ChatGPT
Poll a build started by generate_clip or create_video. Free and instant. Returns state ('planning' while the script, plan and price are prepared — nothing rendered, paid steps wait for approval | 'rendering' | 'awaiting_confirmation' with the cost quote | 'awaiting_review' when a generated artifact needs a look (costs nothing — approving it does NOT approve the budget) | 'awaiting_approval' for an unpriced step | 'awaiting_answer' with the agent's question | 'complete' with video_url and the credits actually spent | 'error' | 'idle' when nothing is running and nothing will on its own, e.g. a restyle waiting for a start in the Synthesys app), scene progress, the output aspect_ratio, and what to do next. Poll every 30-60 seconds — single clips take 8-13 minutes, full productions usually 10-25 minutes (do not give up before 30 minutes). Read-only: polling only reports the status and never starts, resumes or charges anything — a build interrupted by a server restart is resumed by Synthesys on its own, so keep polling.

list_actors

ChatGPT
List the people who can appear on camera: the user's own twins (kind 'my_twin' — their recorded face, and their voice when they cloned it) first, then Synthesys stock actors. Free and fast. Returns actor_id, name and gender for each; stock actors also have a preview_url. A twin is listed by name only (no picture of the user's face is shared) and says which voice it speaks in. Show the user a few and pass the chosen actor_id to create_video. Filter with query (name) and gender; page with limit/offset. Skip this when the user has no preference — Synthesys casts a presenter that fits the brief.

list_brand_kits

ChatGPT
List the account's Brand Kits (logo, colours, fonts, tagline, call to action) and which one is active. Free and instant. Pass a brand_kit_id to create_video to make the video on-brand — it closes on the brand's end card and follows its look. Brand Kit is part of the Pro plan and up; on other plans this returns available:false and videos simply build without one.

list_files

ChatGPT
List the images and videos already in the user's Synthesys library (product shots, logos, footage they uploaded in the app or through an upload link), newest first. Free and instant. Each file has an asset_url — pass it in the reference_asset_urls of create_video or generate_clip. Use this instead of asking for a link when the user says the file is already in Synthesys, and after they use a create_upload_link page. Pages with limit/offset: total counts every match, and next_offset (present when there are more) is the offset of the next page.

list_models

ChatGPT
List the models Synthesys can generate with, for images or for video clips: each model's name and maker, the aspect ratios, resolutions and (video) clip lengths it offers, and its EXACT price in credits per resolution (100 credits = $1). Free and instant. Use it when the user wants a specific model, asks what is available, or wants to compare prices; then pass the model_id to generate_image or generate_clip. Prices are for the settings you ask with (aspect_ratio, duration_seconds for video, images for images), so ask again to re-price. Full video builds (create_video) choose their own models and take no model input.

list_templates

ChatGPT
List every video template (UGC ad, cinematic commercial, faceless story, explainer, …) with its prompt starter, default/max duration, aspect ratio, and available visual styles. Free and instant (<2 s). Call this FIRST — before generate_clip or create_video — to pick a valid template key and keep the requested duration inside the template's clamp.

list_voices

ChatGPT
List the voices that can narrate a video: the user's own cloned voices (kind 'my_voice') first, then the Synthesys voice library. Voices the user made for Studio speak only there, so they are not listed. Free and instant. Returns voice_id, name, style and a preview_url (an audio sample) for each. Pass the chosen voice_id to create_video. Skip this when the user has no preference — Synthesys picks a narrator that fits.

open_in_editor

ChatGPT
Free and instant. Get the editor_url that opens a project in the Synthesys video editor, for the user to review it or keep editing by hand with drag and drop. Works for finished create_video and generate_clip projects and for projects from edit_video. Changes made in either place show up in the other.

recreate_from_url

ChatGPT
Recreate a competitor/reference ad as a brand-adapted video of your own. Pass a public link to the ad — a direct video file (.mp4/.mov/.webm), an ad page with an embedded video, or a TikTok/Meta/YouTube link. Synthesys downloads and analyzes it (verbatim transcript, labeled beats: hook/setup/reveal/proof/CTA, pacing, visual style) and drafts a brand-adapted recreate plan. COST: the analysis is a paid step — 10 credits (~$0.10), charged when it completes (nothing is charged if it fails or is refused). Tell the user before calling, and call get_credits first if the balance may be low. This call is SYNCHRONOUS and slow: ~1-3 minutes for direct video links, longer for page links — do not give up early, and do not re-send the same link while it runs. mode 'page' instead captures a PRODUCT or LANDING page (photos, copy, product facts) and plans an ad for that product; the default 'ad' expects an ad video. Returns the analysis + a project_id, then pauses at a RENDER COST QUOTE (state 'quote_ready' or 'planning' — poll get_video_status until the quote appears). NOTHING is rendered until you approve with create_video {project_id, confirm: true}; the re-render then usually takes 10-25 minutes. Recreates follow the reference clip's real length (capped at 60 s). Prefer this over create_video when the user has a specific existing ad (or product page) to work from; use create_video when they describe what they want from scratch.

upload_reference

ChatGPT
Ingest an image or video from a public URL (product shot, logo, footage) into Synthesys so a build can use it. Free; takes a few seconds (the file is downloaded server-side). Returns an asset_url — pass it in the reference_asset_urls of create_video or generate_clip, or mention it in the brief. This needs a LINK: a file the user attached in the chat is not passed to Synthesys. For a file on their device use create_upload_link; for one already in their Synthesys library use list_files.

Capabilities

Writes

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/synthesys"><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/synthesys)