Hedra

Get a model's input schema and capabilities — call this before using model_specific_parameters to learn which parameters the model accepts. A model that supports several input modes (text-only, start frame, reference images, ...) routes automatically from the inputs you attach.

get_model

Run isolate audio and return its job acknowledgment immediately. Costs money. The card tracks completion automatically; use get_generations only for dependent work or requested status. Do not claim to have seen outputs from this acknowledgment.

isolate_audio

List Hedra generation models. Returns the public v3 catalog: one entry per model with its public id, modality, and description. Use get_model for a model's input schema.

list_models

Find voice IDs for generate_speech or convert_speech. Without query, lists your cloned voices followed by the curated library. With query, searches the shared library by description (your clones are in the listing). Pass a returned voice_id as voice.

list_voices

List the workspaces the user belongs to and which one this connection currently generates in (`is_current`). Generations and their charges land in the current workspace; use set_workspace to move the connection.

list_workspaces

Prepare a short-lived, upload-only credential for the Hedra file picker.

prepare_upload

Find reusable image, video, and audio assets in the user's Hedra library. This searches only the user's library, never the web. Pass asset_ids to fetch known assets, or use query/types/since/until to browse and search. A query matches image and video content by meaning where semantic search is available and always matches literal name, description, and generation-prompt text; results combine both rankings. Without query, results are newest first. Returned asset_id values can be passed directly to generate as {"source": "asset", "asset_id": ...}. Only complete, reusable media assets are returned.

query_assets

Run remove video background and return its job acknowledgment immediately. Costs money. The card tracks completion automatically; use get_generations only for dependent work or requested status. Do not claim to have seen outputs from this acknowledgment.

remove_video_background

Choose which wallet generations in the current workspace debit: "credits" (the workspace credit pool, at credit prices) or "usd" (the prepaid API wallet, at USD prices). Ask the user first, after showing both balances from get_balance. Applies from the NEXT call; estimate/get_generations report prices in the unit of the mode in force. Not available when connected with an API key: set the key's mode where keys are managed.

set_billing_mode

Move this connection to another of the user's workspaces: later calls generate and bill there. Ask the user first — it changes where their money goes, for every OAuth client connected to their account, and applies from the NEXT call (the current call's workspace is fixed). Not available when connected with an API key: a key is pinned to its workspace.

set_workspace

Upload local reference images, audio, or videos directly to Hedra using an embedded file picker in MCP Apps hosts. Use this for files on the user's device; ordinary chat attachments are not automatically accessible. Wait for the user to select and upload files; use the resulting file URLs as media inputs. Clients without UI get a Hedra library link and the multipart endpoint for CLI uploads. Never ask users to host files on a third party.

show_file_upload_widget

Ingest a publicly reachable file URL (image, audio, or video) into Hedra's ephemeral storage and return a URL usable as a generation reference — pass it in the matching *_url or *_urls argument of the generation or media operation tool. The returned URL is the file handle and lasts one hour from upload; the result's `file.expires_at` says exactly when. After that, upload the file again for a fresh handle rather than reusing the old URL. Only public http(s) URLs are accepted.

upload_file

Run upscale and return its job acknowledgment immediately. Costs money. The card tracks completion automatically; use get_generations only for dependent work or requested status. Do not claim to have seen outputs from this acknowledgment.

upscale