Turbogen MCP server
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TURBOGEN_API_KEY | Yes | Turbogen API key (starts with `lk_`), created from the API page in the Turbogen studio. | |
| TURBOGEN_BASE_URL | No | Base URL for the Turbogen API. | https://turbogen.app |
| TURBOGEN_OUTPUT_DIR | No | Directory where generated images, videos and audio are saved. | ~/Downloads/turbogen |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_modelsA | List available generation models with their ids, aspect/voice choices, options and credit prices. kind: optional filter - "image", "video" or "audio". |
| generate_imageA | Generate an image and save it locally. Returns the file path, cost and remaining credits (plus a preview). prompt: what to draw. model: image model id (see list_models(kind="image")), default nano-banana-2. aspect: aspect ratio, e.g. "1:1", "16:9", "9:16". option: resolution/quality option such as "1K", "2K", "4K"; defaults to the model's first (cheapest) option. reference_images: optional list of local PNG/JPEG/WEBP paths used as references. |
| generate_videoA | Generate a video clip and save it locally. Returns the file path, cost and remaining credits. prompt: the shot - subject, action, camera move, light. model: video model id (see list_models(kind="video")), default veo-3.1-fast. aspect: "16:9", "9:16" (some models also "1:1" etc.). option: duration such as "4s", "6s", "8s"; must be one the model offers. image_path: optional local image for image-to-video (first frame / reference). Video takes 1-5 minutes; the call waits up to 10 minutes. |
| generate_audioA | Generate audio (voiceover, background music, a song or a sound effect) and save it locally. text_or_prompt: for "voice" the text to speak; for "music"/"song"/"sfx" a description of the sound. model: "voice" (text to speech), "music" (instrumental background), "song" (sung, needs lyrics) or "sfx". aspect: the voice for "voice" (e.g. "Deep narrator", "Warm female"), the mood for "music" (e.g. "Cinematic", "Lo-fi"), the vocal for "song" ("Female vocal", "Male vocal", "Duet"); defaults to the first choice. See list_models(kind="audio"). option: length/duration, e.g. "300 chars" for voice, "30s" for music, "60s" for song, "5s" for sfx; defaults to the first. lyrics: song lyrics, required for "song". |
| upscale_imageA | Upscale a local image 2x or 4x and save the result next to other outputs. path: local PNG/JPEG/WEBP file. scale: "2x" or "4x". |
| get_balanceA | Show the remaining credit balance of the account behind TURBOGEN_API_KEY. |
| job_statusA | Check a generation job by id (e.g. one that outlived the 10-minute wait). download: if true and the job is done, save its file to the output folder and return the path. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 7 tools
Each tool maps to a distinct action and resource: discovery (list_models), generation per media type (image/video/audio), upscaling, balance check, and async job lookup. The media-type split between generate_image, generate_video, and generate_audio is explicit in the names, so an agent can select correctly without reading descriptions in depth.
Six of seven tools follow a clean snake_case verb_noun pattern (list_models, generate_image, generate_video, generate_audio, upscale_image, get_balance). The outlier is job_status, which drops the verb (a get_job_status form would be fully consistent), but it remains clearly readable.
Seven tools is well-scoped for a media generation server: one per output modality, plus the necessary supporting operations (model catalog, balance, async job status). No tool feels redundant or like filler.
The surface covers discovery, generation for all three modalities, upscaling, cost/balance visibility, and recovery from timed-out jobs via job_status with download. Minor gaps remain: no job listing/cancellation, no image edit/variation operation, and no separate download/fetch tool for arbitrary job ids beyond job_status.