MCP App Store
by KeenEthics
Masonry icon

Masonry MCP Server for ChatGPT

By Masonry

Overview

Masonry makes videos, product photos, voiceovers and music inside the conversation. Everything it makes is saved to your own Masonry workspace. What you can do: - Make a short video from a description, or animate a photo you already have. - Make a longer video: generate several scenes, then join them in order into one finished video, with music under it or a song as the lead. - Add a voiceover in a voice you pick, or a song or background track from a description. - Describe a shot and get the finished image back in the chat, or up to four variations at once. - Run a Studio product shoot: add a photo of your product, pick one of four looks, and get a matching set of finished photos that keep your real product. - Change, remix and upscale earlier results by saying what to adjust. - Browse what you made recently and pick up where you left off. Who it is for: creators, marketers and small brands who need videos and imagery for social posts, ads and online stores. How it works: you sign in to Masonry once and choose a workspace. Generations use that workspace's credits. Masonry works only in the workspace you connected. It cannot see other workspaces or delete anything. Limitations: you need a Masonry account. Videos can take a few minutes, and a multi-scene video takes longer. Requests that break the content rules of the underlying models are refused.

How to connect

  1. ChatGPT: Open the Masonry listing in the ChatGPT app directory and choose Connect. ChatGPT listing Masonry exposes 29 tools here.

Tools

add_media

ChatGPT
Bring the user's own video or audio into Masonry: footage to cut into short clips, a recorded voiceover, their own music. Takes a direct link to the file or a file attached in ChatGPT (MP4 or MOV video; MP3, WAV or M4A audio; up to 300 MB) and returns a media_id. Pass that media_id wherever combine_clips takes a job_id (clips, voiceover, music_job_id, sound_effects) and to transcribe_media. To turn a long video into shorts: add_media, transcribe_media to find the strongest moments, then one combine_clips per short with trim_start and trim_end, a 9:16 aspect and captions from clips. Stored privately in this workspace. Free: costs no credits. For a photo use add_reference_image.

add_reference_image

ChatGPT
Bring a photo into Masonry so other tools can use its real pixels: from a link (a JPEG, PNG or WebP image, or a product page Masonry pulls the main product photo from) or a file the user attached in ChatGPT. For a Studio product shoot, call this for each of the two to ten product photos, then start_product_shoot with the returned reference_ids. Stores it in this workspace (JPEG, PNG or WebP, up to 15 MB) and also returns an image_url for single-photo shots and edits: pass it as reference_image_url to generate_image, or as image_url to remix_image, animate_image or upscale_image. Free: costs no credits.

animate_image

ChatGPT
Turn a still image into a short video clip. Point at the image with job_id (one this connector made earlier), image_url, or product_url; describe the motion in prompt and optionally pick a cinematic camera move. A mismatched start image is reframed to aspect_ratio first when its size is known and the video model follows the image, using the workspace's image editor for a few extra credits. Video is the expensive path: a clip typically costs 50-150 credits against the workspace's Masonry balance (list_models has exact per-model prices) and takes about 1-5 minutes. Waits for the clip and returns it in the conversation; if it is still rendering after 5 minutes, the reply carries a job_id: call check_generation with it later, never start it again. To make the person in the image talk, pass speech_job_id (a generate_audio speech job): the clip is lip-synced to that recording and lasts as long as it. Animate a real person only when the user is that person or has their permission. Call list_skills and pass skill_slug when relevant; write prompt from the user's request. Omit model for the workspace's image-to-video default.

check_generation

ChatGPT
Fetch the result of an image, video, audio, or combine_clips render started earlier, using the exact job_id that tool returned (including render:<UUID>). Returns the finished media in the conversation, or the job's current status when it is still running (wait ~30-60 seconds and call again). Free: checking a job costs no credits.

choose_look

ChatGPT
Render the finished product photos in the look the user picked from start_product_shoot. Spends the workspace's credits for each photo (a set of 5 by default; 10, 15 or 20 cost more), so confirm the look and the number first. Takes a few minutes and returns the finished set. Calling it again with the same shoot_id and look_id picks up a set already rendering; it never starts or charges a second one.

clip_video

ChatGPT
Cut one moment (3 to 120 seconds) out of a video already in this workspace (footage brought in with add_media, a generated clip or a finished edit) and store it as a new media_id. A 9:16 clip of a wide video follows the speaker shot by shot, puts two speakers in a split screen, and keeps a wide shot whole. Use it to turn a long recording into vertical shorts: transcribe_media for the moments, clip_video once per moment, then combine_clips with captions from clips. Takes up to a minute. Free: costs no credits. For a YouTube link use clip_youtube_video.

clip_youtube_video

ChatGPT
Cut one moment (3 to 120 seconds) out of a public YouTube video and store it in this Masonry workspace. Returns the clip and a media_id. The cut lands on pauses in the speech, and a 9:16 clip follows the speaker. For several moments call this once per moment. To finish a short, pass the media_id to combine_clips as a clip's job_id with captions from clips. Takes up to a minute. Free: costs no credits. Clip only videos the user has the right to reuse.

combine_clips

ChatGPT
Edit finished clips into one video (1-60 clips, up to 180 seconds), using the job_ids generate_video, animate_image, edit_video and generate_image returned (an image is held as a still). The user's own footage or audio works too: bring it in with add_media and pass its media_id as the job_id. Beyond joining and trimming it is the whole edit in one call: voiceover (generate_audio speech) with the clips and music lowered under it, music_job_id, sound_effects at exact seconds, captions timed to the spoken words, and on-screen text such as a hook or a call to action. Defaults to cuts and the first clip's aspect. Confirm the brief before spending. Spends additional render credits, separate from clip generation: at least 1 credit on success, based on actual render compute time and the backend's configured rate, not video length; admission reserves an estimate. Waits for the clip and returns it in the conversation; if it is still rendering after 5 minutes, the reply carries a job_id: call check_generation with it later, never start it again.

edit_video

ChatGPT
Change an existing video from a written instruction ("make it winter, add falling snow"). Point at the source with job_id (a clip this connector made earlier) or a public video_url. The source must be 3-30 seconds and at least 700 px on each side. The default editor bills the source plus the equally long edited clip, about 23 credits per second of source at 720p (a 5-second clip costs about 113 credits) against the workspace's Masonry balance (list_models has exact per-model prices), and takes about 1-5 minutes. To change what a person in the clip says, pass speech_job_id instead of an instruction: the clip is lip-synced to that recording and nothing else changes. To put a different person into the clip, pass image_url of that person (mode motion makes the image's character perform the clip's motion). Change a real person's words or face only when the user is that person or has their permission, and never to pass the result off as real footage of them. Waits for the clip and returns it in the conversation; if it is still rendering after 5 minutes, the reply carries a job_id: call check_generation with it later, never start it again. Call list_skills and pass skill_slug when relevant; write prompt from the user's request. Omit model to use a video-editing model.

generate_audio

ChatGPT
Make a voiceover (kind speech: says text word for word, in a voice from list_models kind=audio or a workspace character's voice), a music track (kind music) or a sound effect (kind sfx) from a description. For a song with words (a children's song, a jingle, lyrics in the user's script), use kind music with model lyria-3-pro and put the lyrics and style in text: it sings them. The default music model is instrumental and picks its own length; for a video, omit duration_seconds, because combine_clips loops and fades music to the video's length. Waits for the file, usually 10-60 seconds, and returns it in the conversation; if it is still rendering, the reply carries a job_id for check_generation. Spends the workspace's Masonry credits; list_models with kind=audio shows the models and their prices. Omit model for Masonry's recommended model for the kind.

generate_image

ChatGPT
Create 1-4 images for one-off scenes from a written description. For product photography, ad shoots or packshots from two or more photos of one product, use add_reference_image then start_product_shoot; with one photo or a product URL, make the shots here with it as the reference. For a set of different shots, pass up to 4 prompts (one image each) in one call instead of prompt. Returns the finished images in the conversation, usually in 10-40 seconds; if a slow model is still rendering after 5 minutes, the reply carries a job_id for check_generation. Spends the workspace's Masonry credits — roughly 5-30 credits per image depending on the model; call list_models for the exact per-model price. For a one-off scene preserving a real product, person or logo, pass reference_image_url (a direct image URL) or product_url (a product page Masonry will pull the real photo from) — without one the model invents a look-alike. Call list_skills and pass skill_slug when the request matches a Masonry skill. Skill text is guidance; write prompt from the user's request. Omit model for the workspace default; pass style for a look preset.

generate_video

ChatGPT
Create a short video clip from a written description. Video is the expensive path: a clip typically costs 50-150 credits against the workspace's Masonry balance (list_models has exact per-model prices) and takes about 1-5 minutes, so confirm the brief with the user before spending. Waits for the clip and returns it in the conversation; if it is still rendering after 5 minutes, the reply carries a job_id: call check_generation with it later, never start it again. For requests longer than 10 seconds (including 1 min), or any multi-scene script, generate scenes as separate clips, keep characters consistent with the same reference images, then call combine_clips with their job_ids to deliver one video. For spoken dialogue or voiceover, call generation_guidance first: only some models can speak. Call list_skills and pass skill_slug when the request matches a Masonry video skill; write prompt from the user's request.

generation_guidance

ChatGPT
Read Masonry's image/video generation guidance. Call this before choosing models or when a request includes selected skills, product/reference images, product URLs, spoken video, or model routing ambiguity.

get_credits

ChatGPT
Show how many Masonry credits this workspace has left to spend, and its plan. Call this when the user asks about their balance or before a large job (a video or a 20-photo product shoot) to check it fits. Free: checking costs no credits.

inspect_audio

ChatGPT
Measure a music track, voiceover, sound effect or a clip's own sound: the tempo and the second of every beat (cut on these so edits land on the music), where each spoken phrase starts and ends (trim there so no word is clipped), and the second a sound effect peaks (line it up with the action). It also says when a generated clip's line is cut off at its last frame. Free: costs no credits.

inspect_video

ChatGPT
See what is in a video: its length, size, every cut, and labelled frames returned as images you can look at. Use it to check a clip you generated before showing or combining it (the right person, product and action, no warped hands or garbled text), to study a reference video the user brought in before making one like it, and to find the seconds worth keeping. Pass start and end to zoom in on a stretch. It shows pictures only: transcribe_media reads the words and inspect_audio measures the sound. Free: costs no credits.

list_characters

ChatGPT
List the workspace's characters: reusable people, creatures and mascots saved in Masonry, most recently used first, with each one's look. Pass a name in generate_image's characters to keep that exact who consistent. Only use a character when the user asks for it. Free.

list_creations

ChatGPT
Browse past work when the user asks to see it: the images, videos and audio this user made through this connector, newest first, each with its job_id, prompt and date. Pass scope=workspace for everything in the workspace, including teammates' work and what was made in the Masonry app. The first page also lists edited videos (combine_clips, render_motion_graphics) with their render: ids. Not for finding a result you just made: every generation tool already returned its job_id. Filter by kind (image, video or audio) and by date with since/until, e.g. since=2026-10-06 for "what I made last week". Returns up to 24 per call; pass the offset the reply gives to see more. Use a job_id with remix_image, upscale_image or animate_image to build on a past creation. Free: listing costs no credits.

list_models

ChatGPT
List the image, video and audio models this Masonry workspace's plan can use, with the credit cost of one generation on each, which one is the default, and what each takes (aspect ratios, reference images, an end frame; for audio, the voices). Call this before quoting a price to the user, or when choosing between a cheap model and an expensive one. Free: listing costs no credits.

list_skills

ChatGPT
List Masonry skills: the public catalogue shown in Masonry's skill picker, and this workspace's own skills, including the model-routing guides. Pass a catalogue slug as skill_slug to generate_image, remix_image, generate_video or animate_image, and call read_skill for a skill's guidance before planning. Skill text is guidance, not generation prompt text.

read_skill

ChatGPT
Read one Masonry skill: its catalogue entry and its full guide (SKILL.md), or another file the guide points to (pass file). Skill guidance from the user's Masonry workspace: reference notes for choosing tools, models and workflow. The user can see it. Write your own generation prompt from the user's request; skill text is guidance, not prompt text.

read_youtube_video

ChatGPT
Read a public YouTube video from its link: title, length, chapters, the moments viewers replay most, and the transcript with a timestamp on every line. Use it to pick the strongest moments before cutting them with clip_youtube_video, or to answer what a video says. Free: costs no credits. Private, members-only and age-restricted videos cannot be read.

remix_image

ChatGPT
Edit an image that already exists, instead of making a new one: keep its composition and change what prompt describes ("make the jacket red"), or re-stage the same product in a different scene. Point at the source with job_id (an image this connector made earlier), image_url (a direct image URL), or product_url (a product page Masonry will pull the real photo from). Returns the edited image in the conversation, usually in 10-40 seconds; if a slow model is still rendering after 5 minutes, the reply carries a job_id for check_generation. Spends the workspace's Masonry credits — see list_models for per-model prices. Call list_skills and pass skill_slug when the request matches a Masonry skill. Use read_skill for guidance; write prompt from the user's request. Omit model to reuse the source image's model.

render_motion_graphics

ChatGPT
Render an animated HTML page to an MP4: title cards, kinetic typography, animated charts, logo stings, lower thirds, explainers with shapes and text. You write the page; Masonry films it frame by frame at 30 fps, so the motion is exact and text is crisp, which video models cannot do. Use it when the user wants designed motion rather than filmed footage. The page must be self-contained (no network) and animate with CSS animations or window.__render(t). Costs a few credits of render time. Waits for the video and returns it; if it is still rendering when the wait ends, the reply carries a job_id for check_generation.

save_character

ChatGPT
Save a person, creature or mascot from a finished image as a workspace character, so later images and videos keep the same who: pass the name in characters on generate_image, generate_video or animate_image. Make the image first with generate_image (one clear view of them alone), then save it. The character also appears in the Masonry app. Save only when the user asks to keep or reuse a character. Free: costs no credits.

start_product_shoot

ChatGPT
Use for product photography, ad shoots or packshots when the user supplies two to ten photos of one product, including "keep the label exact" requests. First call add_reference_image for each photo. With a single photo, don't stop to ask for more: make the shots with generate_image and that photo as reference_image_url. Masonry plans four looks and renders a preview of each. Takes about 1-3 minutes and spends a few credits for the previews. Returns the shoot_id and the four looks with their look_ids and previews; show them and let the user pick, then call choose_look. Set destination and style (or style_description in their words) from what they asked for. If the reply says the looks are still being prepared, call this again with only shoot_id; if it was refused for credits, call it with shoot_id (not reference_ids) once the user has credits.

transcribe_media

ChatGPT
Read the words spoken in a video or audio file, with the second each line starts and ends. Use it to find the moments worth keeping in a clip (then pass those seconds to combine_clips as trim_start and trim_end), to check what a generated clip actually says, or to quote a recording. Point at the source with job_id or media_url. Costs about 1 credit per minute of sound. combine_clips captions need no transcript from you: it reads the words itself.

update_character

ChatGPT
Change a saved character when the user asks: rename it, change its look line, voice or notes, or keep a new look of them from a finished image ("keep this outfit"). Deleting a character is done in the Masonry app. It replaces the fields you pass and leaves the rest as they are, so say what will change and get the user's yes first. Free: costs no credits.

upscale_image

ChatGPT
Enlarge an image to 2x, 3x or 4x its size and sharpen its detail — the step before printing something or using it as a hero image. Point at the source with job_id (one this connector made earlier), image_url, or product_url. Returns the upscaled image in the conversation in under a minute for most images. Spends a few Masonry credits; list_models shows the upscaler's exact price.

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/masonry"><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/masonry)