MCP App Store
by KeenEthics
Tesseract icon

Tesseract MCP Server for ChatGPT

By Mirage

Overview

Tesseract brings professional creative tools into your agent workflow. Work with video layers, compositions, keyframes, and audio in one connected project. Preview the result, direct a revision, and render the finished video locally.

How to connect

  1. ChatGPT: Open the Tesseract listing in the ChatGPT app directory and choose Connect. ChatGPT listing Tesseract exposes 12 tools here.

Tools

generate_image

ChatGPT
Start an image generation from a prompt. Returns generation_id, model_id, modality, and status; pass them to poll_generation to get the signed download URL and asset_id. The completed image may be attached to the poll result.

generate_music

ChatGPT
Generate music. ElevenLabs requires prompt with duration_ms, or composition_plan alone. Sonilo text-to-music takes prompt and optional duration_ms (default 90000). Sonilo video-to-music requires video_url and duration_ms; prompt is optional. start_offset_ms and prompt_influence apply only to video-to-music. demucs-voice-removal takes audio_url (audio or video, at most 10 minutes and 100 MB) and returns the same sound without the voice; demucs_model, shifts 1-10, and output_format mp3/wav apply only to it. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the audio URL and asset_id.

generate_sound_effect

ChatGPT
Start a sound effect or sound sequence from a prompt. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the MP3 URL and asset_id.

generate_speech

ChatGPT
Speak text verbatim. google-gemini-3.8-flash-tts requires a Captions voice_id and accepts style_prompt. elevenlabs-v4 takes a premade voice name, supports audio tags such as [whispering] and IPA pronunciation, and returns an MP3. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the signed audio URL and asset_id.

generate_video

ChatGPT
Generate video from a prompt, frames, or reference media. Match the model to the inputs: text-, image-, or reference-to-video. Recast (minimax-h3-max-recast) replaces the people of a source video: pass video_url (5-30 s, no shot over 15 s) with 1-4 reference_image_urls (left to right); it bills $0.30/s at 768p or $0.45/s at 1080p. Seedance supports generate_audio; Seedance Fast also supports bitrate_mode; MiniMax supports prompt_expansion_mode. Frames and references need direct HTTPS URLs; reference audio needs an image or video. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the signed video URL and asset_id.

list_skills

ChatGPT
When the user names a Mirage recipe, or asks for a video of a kind Mirage serves a recipe for, read the matching served skill before generating anything. When the served skills include the core skill (the one that describes the shared generation workflow), follow it and the matching skill through the finished edit; if it is not served, continue with the matching skill and do not retry. Other generation requests need no skill. ad-localizer: The same ad in another language: every on-screen word translated in place, speech re-voiced, music kept. Use when: They want their own ad or product video translated, localized, dubbed or adapted for another market. Not someone else's ad (Ad Remix) or a new ad (Product Ad) ad-remix: An ad they like, remade for their product: its structure and pacing, new product shots and music. Use when: They share an ad and want "this, but for my product", or want to remix, recreate or borrow an ad's structure car-storytime: A creator telling a story to their phone from a parked car, with their photos overlaid. Use when: A car vlog, car storytime, "talking in my car" or car confessional. Not a polished coaching talk (Coach Explainer) or a product demo (Creator Cam) cartoon-explainer: An animated explainer on any topic: one cartoon character, narrated, with music. Use when: A cartoon, animated, stickman, whiteboard or classroom explainer of a topic. Not a presenter explaining to camera (Explainer), their footage as a cartoon (Footage to Cartoon) or a zombie what-if (Zombie Explainer) creator-cam: A UGC selfie video (review, tip, testimonial, hot take) with jump cuts, captions and music. Use when: UGC, a UGC ad, an influencer or creator video, a product review or a talking selfie video. Not a try-on, unboxing or car vlog day-in-my-life: A vlog of a day, with captions or a light voiceover and music. Use when: A day in my life, a vlog, a morning or night routine, "come with me", or a day in a job's or brand's life. Not one talking selfie (Creator Cam) earth-zoom: One continuous dive from space onto the first frame of their video. Use when: An earth, world, Google Earth, satellite or zoom-from-space intro onto their footage. Not other hooks (Hook Generator) explainer: A presenter, a coach or expert, explaining or giving advice to camera, with b-roll, key words and music. Use when: An explainer ("explainer on …", the older "coach explainer"), advice, a tip, a story or the news told by a presenter to camera (business, finance, fitness, mindset, sports), or a podcast-style clip. A cartoon or animated explainer is Cartoon Explainer, a zombie or survival what-if is Zombie Explainer; not a casual selfie video (Creator Cam) fit-check: A cinematic outfit video cut on the beat with bold one-word titles, no talking. Use when: A fit check, OOTD, outfit, lookbook or clothing-drop video of one look. Not several pieces held up and worn by a creator (Try-On Haul) footage-to-cartoon: Their footage redrawn as a cartoon, on the clip's own timing. Use when: They want their video cartoonified, turned into anime or animation, or "make this a cartoon". Not a topic explained from scratch (Cartoon Explainer) hook-generator: A 1–3 second scroll-stopper made from their opening frame, with an impact sound, cut in front of their video. Use when: A hook, an intro, a scroll-stopper, a pattern interrupt, or new openings to test. Not a dive from space (Earth Zoom) or a whole ad music-video: A sung performance video: the song first, then the performer across 2 or 3 places. Use when: A music video, a performance or live-session clip, or someone singing their song. Not music added under a video they have (generate) product-ad: A short promo for a product or app: product shots, kinetic words, an end card, music. Use when: A product ad, promo, launch or app video, or product b-roll. Not a creator talking about it (Creator Cam), a remake of an ad (Ad Remix) or a translation (Ad Localizer) real-estate: A listing video with an agent presenting the home over a walkthrough. Use when: A realtor's listing video, property tour, walkthrough, just-lis…

poll_generation

ChatGPT
Check a generation once. Pass the generation_id, model_id, and modality returned by generate_* or transcribe_audio. SUBMITTED, IN-PROGRESS, and FINALIZING are still running; poll again later. COMPLETE returns data (a fresh signed download URL for media or a 3D model, or transcription data) and asset_id when available. FAILED returns failure details in data.

read_skill

ChatGPT
Read one served skill by skill_id. Without path, returns its SKILL.md and file list. With path, returns that file: text as text, or binary as an embedded resource with exact bytes, skill:// URI, MIME type, size, and SHA-256 digest. Anonymous text reads are limited to 64 KiB; larger text and all binary reads require sign-in. Images up to 1 MiB are also attached. Call list_skills first to find a skill_id or file path.

segment_image

ChatGPT
Start a SAM 3 segmentation operation on a source image and return one stored mask. Requires image_url (PNG, JPEG, WebP, or GIF; up to 30 MB and 8192 px per side) and at least one of prompt (text such as 'wheel'), point_prompts, or box_prompts. Coordinates are source-image pixels (0,0 is the top-left edge); label 1 points at the object (default) and 0 at the background, and prompts with the same object_id refine the same object. SAM 3 returns exactly one mask: apply_mask=true (default) cuts the object out on a transparent background; false returns the binary mask. Multiple masks, scores, and boxes are not supported. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the mask's signed URL and asset_id.

segment_video

ChatGPT
Start a SAM 3 segmentation/tracking operation on a source video and return a rendered MP4 or WebM. Requires video_url and at least one of prompt, point_prompts, or box_prompts. Coordinates are normalized 0-1; label 1 selects and 0 excludes, and prompts with the same object_id refine the same object. The source may be MP4, MOV, WebM, or M4V, up to 200 MB and 18,000 frames. Returns generation_id, model_id, modality, and status; pass them to poll_generation for the rendered video's signed URL, asset_id, and resource link. The provider's box-prompt ZIP export is not returned.

tesseract_viewer

ChatGPT
Open and preview a .tsrct file.

transcribe_audio

ChatGPT
Start transcription from a publicly reachable audio URL. Returns generation_id, model_id, modality, and status; pass them to poll_generation for data.text and word-level timestamps in data.words. Omit language_code to autodetect.

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/tesseract"><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/tesseract)