Astria MCP server

Use this when choosing an Astria image, video or music model or checking supported media inputs, image resolutions, aspect ratios, lyrics limits, song prices, and defaults. Pass type (image, video or music) to get only that catalog. Returns the live model catalog and model names/tune IDs; does not generate media or spend credits. For video generation, submit null model and resolution fields unless the user explicitly selected them; let Astria resolve defaults. For images and music, use the explicit user selection or catalog default.

Use this when finding an Astria workspace for a team or brand photoshoot. Returns authorized workspace IDs; expand migration to include their website URLs for duplicate detection. Use the selected ID consistently for references, templates, and generations. Does not create a workspace or generate media.

Use this when finding a saved person, model, garment, product, or SKU for consistent photos, virtual try-on, headshots, or video. Search title/product name/SKU; gallery=true finds public model references. Returns reusable reference IDs and subject classes. Does not create references or generate media; inspect a selected match with get_reference.

Use this when inspecting a selected Astria identity or product reference and its training images before a photoshoot or try-on. Returns the saved subject class and reference details. A reference token uses <model_type:ID:1> followed by that class. Does not create or alter the reference.

Update a selected reference title, subject class, short description, injected text or verified catalog source records. text sets characteristics.text; omit it to preserve the current text or pass an empty string to clear it. Other characteristics are preserved. Sources include provider, account_key, resource_type, external_id and captured metadata. Does not retrain the reference or generate images.

Use this when a supplied person or product needs one reusable Astria reference for consistent headshots, virtual try-on, product shots, or multiple scenes. Provide real HTTPS photo URLs, a subject class in name, and a human label in title. Reuse an existing reference when available. Creates paid content only within the requested generation/reference scope; requires an idempotency_key. Returns the new reference ID and class; reuse them across generations. Does not return a finished photoshoot.

Use this when the user asks to create or edit images: product photography, ecommerce packshots, professional headshots from supplied photos, or consistent people/products across scenes. For on-model fashion, virtual try-on, and lookbooks, use generate_lookbook_images instead. Reference tokens in text preserve selected subjects; input_image edits an existing HTTPS image, and mask_image limits an edit. Separate scene briefs need separate submissions; num_images varies one brief. model is optional: omit it for the catalog default image model, and call list_models type=image only to choose another model or check resolutions. Respect an explicitly chosen provider. Prompt advice or a text-only storyboard is not a generation request. Creates paid Astria content within the requested scope; use a unique idempotency_key per action and reuse it only for an identical retry. Returns a submitted prompt with id and tune_id and an embedded result preview; check get_prompt for finished assets or error…

Generate on-model fashion, virtual try-on, and lookbook images using separate avatar and product references. Before submitting, visually inspect EVERY reference image, including the training images of saved references retrieved with get_reference. Reading reference names or metadata alone is not inspection. Avatar references MUST be either head-and-shoulders portraits without a visible outfit, or body images wearing plain, neutral, close-fitting clothing such as a solid-color swimsuit or skin-tone bodysuit. Garment and accessory references MUST be product-only packshots without a person wearing them. Do not use dressed lifestyle or editorial photographs as avatar or product references. If references do not qualify, prepare suitable references within the user's requested scope or request replacements before generating. Use the avatar for identity and the packshots for the outfit. Reference tokens in text preserve selected subjects; input_image edits an existing HTTPS image, and mask_im…

Use this when the user requests an Astria video, animating a photo/product, or motion with consistent person/product references. Provide video_prompt. Submit video_model=null and video_resolution=null unless the user explicitly selected those settings; Astria resolves its defaults. Do not select a model or resolution yourself or copy catalog defaults into the request. Call list_models type=video only to resolve an explicit selection or check capabilities. generate_audio=true picks the audio output of families that offer one. Use first_frame/last_frame for image-led video and video_references/audio_references only when the chosen model supports them; the prompt names their items by position as @Video1 and @Audio1. Do not call for storyboard writing, prompt advice, or an explicitly chosen other provider. Creates paid content within the requested scope; use a unique idempotency_key per action and reuse it only for an identical retry. Returns a submitted prompt with id and tune_id and an…

Use this when the user asks for an Astria song, jingle, soundtrack, background music or instrumental track. Provide text as the song description (genre, mood, instruments, vocals, tempo); pass lyrics with the model's section tags such as [Verse] and [Chorus] to sing given words, omit lyrics to let the model write them, or set instrumental=true for no vocals. model is optional and defaults to default_music_model; call list_models type=music to choose another music model or check lyrics length limits, section tags and the price of each kind of song. For MiniMax, audio_reference selects Music Cover: an MP3 with vocals, 6 seconds to 6 minutes, a 10-300 character style prompt, and no lyrics or instrumental flag. It does not guarantee preservation of exact event timing. Do not call for lyric writing or music advice alone; use generate_video generate_audio for native video sound. Creates paid content within the requested scope; use a unique idempotency_key per action and reuse it only for an…

Use this when finding recent Astria images/videos, checking existing submissions after an uncertain response, or locating outputs from a saved photoshoot run. Filter by reference, template, order, text, review status, source, or media type. Does not generate or charge credits. Results include prompt IDs for get_prompt; expand prompt.comments to inspect review feedback.

Use this when checking progress, finished images/videos, or errors for an existing Astria generation. Pass its returned id. Returns the current prompt and available assets with an embedded preview; no generation is submitted or charged. Pending or failed results must not be treated as a new generation request. expand prompt.comments includes image review feedback.

Assign an existing prompt to a template, clear template membership, or move it to an authorized workspace. Pass pack_id or base_pack_id as the string null to clear that assignment. Does not generate, delete or retry an image.

Use this when the user asks to upscale a finished Astria video or increase its frame rate. Pass the prompt id of a completed video and either video_upscale_model or video_fps_model, not both; for both, call again with the first result id once it completes. Creates a new paid prompt and leaves the original unchanged. Each enhancement applies once per video. Use a unique idempotency_key per action and reuse it only for an identical retry. Returns the submitted prompt with id and cost_mc; check get_prompt for the final video or errors without resubmitting.

Append an uploaded image or MP4 to an existing prompt, for example a finished muxed film. First prepare an upload with create_upload and successfully upload its bytes, then pass its blob_signed_id and the prompt id. Supports PNG, JPEG, WebP, AVIF and MP4. Preserves existing results; does not generate or replace media. Requires an idempotency_key. Returns the saved asset and updated prompt; use get_prompt to view all results.

Replace a saved image or MP4 result only when requested. First successfully upload the replacement with create_upload, then pass its blob_signed_id and the prompt id. Omit replace_blob_id only when the prompt has exactly one result of the replacement media type; otherwise select its blob_id from get_prompt. Retains the original in debug history and preserves other results. Does not generate media. Requires an idempotency_key; reuse the same key and arguments for a retry. Returns the saved asset and updated prompt; use get_prompt with expand=["debug_images"] to inspect history.

Use this when finding an existing reusable Astria photoshoot/template (also called a pack) for product images, lookbooks, or headshots. Returns accessible template names/slugs and thumbnails, most recently used first; inspect a selected template with get_template before run_template. Browsing does not run a photoshoot or spend credits.

Use this when resolving a selected Astria photoshoot template by slug or ID before running it or inspecting its prompts, reference slots, and baseline pricing. Does not generate images or modify the template. Preserve the returned reference-slot order when calling run_template.

Use this when reading what a template's prompts say, to learn from or rebuild them: each prompt's verbatim text, aspect ratio, model, references and example images, as plain text with one section per product class. A reply holds at most 50 prompts and counts the sections; read further with offset only when the user asks for more. Does not generate images or modify the template. Use get_template for its reference slots before run_template.

Create a reusable photoshoot template container in the selected workspace. Supply a title, model and optional main product category; author its shots with generate_images using the returned pack_id, then set reference categories with update_template. Creating the container does not generate images.

Update an existing template title, subtitle or reference categories. Does not generate images or replace its prompts. Inspect get_template before changing its category slots.

Create a separate brand/team workspace when requested. Supply its title and optional canonical website URL. Returns the workspace ID; use it on subsequent references, templates and generations. Does not invite members or generate images.

Attach images to a selected workspace before using them on its landing page. Supply image_urls for public HTTPS images (including completed Astria direct uploads), signed_ids from completed create_upload uploads, or both. Reuses Astria uploads and imports external images into workspace storage. Supports PNG, JPEG, WebP, AVIF and GIF up to 100 MB each. Preserves existing workspace images and does not generate media. Returns workspace_id and images with id, blob_id, url, filename and content_type; use the returned URLs in landing_page_html. Requires an idempotency_key for safe retries.

Inspect a selected workspace by ID, including its brand website, landing_page_html and attached images with persistent URLs. Does not create or change the workspace.

Update a selected workspace title, brand website URL or supplied landing page HTML. Does not publish the workspace, invite members or change billing. Provide only the fields to change. Set brief only when the user requests full landing-page regeneration; it replaces existing HTML. Use get_workspace to read the generated HTML. Requires an idempotency_key for safe retries.

Use this when the user asks to run a saved Astria photoshoot/template on their person or product, repeat an existing lookbook with a new SKU, or generate portraits from a selected pack. Resolve the template and reference slots first with get_template; pass tune_id entries in slot order, supported photo inputs, or tune_replacements mapping original IDs to replacement ID arrays to swap selected products while retaining other references. Do not substitute one generic image for a requested template run, and do not run a template for browsing or pricing advice. Creates paid Astria content within the requested scope; requires an idempotency_key. prompt_attributes supplies additional per-prompt overrides as a JSON object of string values and takes precedence over matching named parameters. Returns the submitted order/prompt IDs; use list_prompts to find prompt IDs and get_prompt to check outputs without rerunning.

Use this to follow a run started with start_agent_run; poll every 15-30 seconds while status.state is submitted or working. Returns the task: status.state (submitted, working, input_required, completed, failed or canceled), status.message with the agent's latest words, pending questions, artifacts and an Astria url. On input_required, put the questions to the user and answer them with reply_agent_run. Artifacts are the results: a file part has a downloadable uri (such as the final MP4) and a data part has a prompt_id for get_prompt. Does not start or charge anything.

Use this when get_agent_run reports input_required. Pass the run id, the question's tool_call_id and answers mapping each question text to the chosen option label (an array of labels for multi-select, or the user's own words). The run resumes and may generate paid media within its original request; requires an idempotency_key. Then poll get_agent_run; the task can still show the answered question for a few seconds.

Use this only when the user asks to stop a run started with start_agent_run. Interrupts the working agent; generations it already submitted still finish and are billed. Only the user who started the run can cancel it. Requires an idempotency_key. Returns the task; get_agent_run reports canceled once the run stops.

Refresh one embedded Astria generation. Presentation only; does not generate or charge credits.

Check complete present_generation-shaped drafts before generating or presenting them. Returns valid, field-specific errors with allowed_values and repair guidance, and unchecked submission checks. tune_id is the numeric image or music model id and video_model a family id from list_models. Models are checked against the composer catalog of the selected workspace; page-only context, such as a fine-tune page opening its own base model, is unavailable, so set storyboard yourself for /storyboard requests. File availability, media metadata, account permissions and reference tune access are not checked. Does not create prompts, upload files or charge credits. Fix every error and validate again before submitting.

Discover Astria creative workflows, including music videos, motion design and soundtracks. Returns skill names, trigger descriptions and metadata without their instructions. Use get_skill to read a selected workflow before executing it. Reading does not generate media or spend credits.

Read the current Astria workflow instructions by name, including an explicit /skill-name request. Returns the full Markdown directly; no shell or file access is needed. Read named dependencies with this tool too. Execute the workflow directly with available Astria tools and local capabilities. Reading a skill or a multi-step request does not require start_agent_run; delegate only on explicit user request or for a required step the current host cannot execute, limiting the handoff to that stage when practical. Assess actual tools rather than desktop, web, or embedded client labels. When executing a database workflow, pass its name as skill_name on generation tools and present_generation drafts, or --skill-name in the CLI. Keep the main workflow name across helper skill reads, but a generation that tests or runs a saved template skill, such as create-lookbook testing lookbook-acme-girls, takes the name of that template skill; omit it for unrelated work. Reading does not start a run or g…

Save reusable Markdown instructions as a private personal or workspace skill, owned by the connected user. Workspace members can read and edit workspace skills. Names are globally unique lowercase slugs. This does not publish a public skill or execute its instructions.

Edit a personal skill you own or a skill in an accessible workspace. Supply its existing name and any description, body or metadata changes. Omitted fields are preserved. Public source-managed skills cannot be edited through this tool; create a new private skill with a distinct name instead.

Refresh one embedded Astria agent run. Presentation only; does not start, answer or charge anything.