Skip to main content
These are the tool names exposed by Aspect’s remote MCP server. Your client may display a server prefix in front of them. Inspect the connected tool schema for exact input types, required fields, and supported options before constructing a call. Batch IDs when a tool accepts a list. For listing and search tools that expose pagination, request additional pages rather than assuming the first response is the complete library. The list tools for projects, people and their project access, directories, collections, metadata attributes, share links, canvases, and skills return the rows that fit in one response: when it includes truncated: true, call again with offset increased by returned. A tool that works on a list of items processes each item on its own and returns one object. When some items fail, it adds failed, the number of items that failed, and failures, each with the item’s index in your list and why it failed. Both are left out when every item succeeds.

Workspaces

Projects

For files and folders at the top level of a project, call projects_get and use its root_directory_id as the parent directory ID.

Directories

Copies can go to another project in the same workspace. When copying across projects, values of project-specific metadata fields and matches to project-specific people or objects do not carry over. Destination name conflicts are resolved by adding a suffix to the copy’s name.

Assets

Timecode conversion accepts multiple positions for one asset in a single call. If the asset has no frame rate usable for timecode, the tool returns null timecodes with a reason. Use assets_search_capabilities before building metadata, object, or People filters.

Transcription

Asset content

Content and transcript tools return indexed information. assets_content_get returns no entries when the asset has not been processed or has no content for the requested track.

Collections

Metadata

Comments

Comment creation does not support drawing annotations, text highlights, mentions, or attachments. Created and listed links include a client_url. Asset share links also include an embed_url for embedding the asset in an external site.

Other

Canvases

Read canvas_content_get before changing a canvas. Use the returned IDs and preserve existing content. Node positions affect the order of inputs consumed by connected generation nodes. Use frame_slot to target a video’s first_frame or last_frame; an image connected to an Auto node without a frame slot is a reference image. Model input rules still apply to connections created through MCP. Creating nodes does not start generation. See Canvases and generation for node types, connection examples, and running steps in dependency order.

Generation

Pass exactly one of auto_tier (faster or better) or a catalog model_id. Auto chooses a compatible model and adjusts settings to supported values; the run records its actual model and parameters. Use generations_models_list when choosing a specific model or inspecting available capabilities. Starting a generation uses workspace credits and returns a runs list before the work finishes. Call generations_run_wait with those run IDs and call it again while it reports finished: false; each call accepts up to 10 IDs and returns within its timeout (1–50 seconds). When finished, check every run’s state: completed media runs return output_asset_id, text runs return output_text, and failed runs return error. Use generations_run_get for a one-off status check. A run does not need a canvas. For a one-off Auto generation, pass project_id, auto_tier, and modality, plus a prompt and any asset inputs. Media output is committed to the project’s Aspect Stash folder; text stays on the run. For a canvas run, pass canvas_id and node_id, the node’s prompt, and its model choice. The server derives inputs from connected nodes, so do not pass explicit inputs or modality. Follow the generation guide to retrieve the result or download its file.

Skills

Instructions can refer only to the skill’s project and items in it, written @[name](type:id) such as @[Logo.png](asset:<asset_id>), and each reference shows as the item’s current name. A reference to something you can’t open, or outside the skill’s project, reads @[unavailable](type:id); leave it as written to keep it. body in skills_update replaces the whole instructions, so start from skills_get. Aspect skills are read-only. skills_create and skills_update name the skill to load first, which covers how to write one.

File and editor boundaries

MCP organizes cloud resources. Use the Aspect CLI for uploads, downloads, mounts, and local pins. Use the editor’s own integration to operate a timeline. fs_items_copy returns job_id and items_total; it queues a copy instead of waiting for every asset. It accepts include_comments, include_metadata, and copy_as_proxy; the last one matches the Copy videos as proxies option in the Copy dialog. See the agentic editing workflow for a proxy-copy example. The server does not expose workspace creation or deletion, project deletion, emptying trash, storage tokens, or public share sessions. Direct downloads are limited to the short-lived signed URLs from assets_download_url_get, intended as a fallback when the CLI is unavailable. Asset and directory deletion moves those items to project trash. Canvas and collection deletion have their own access requirements and should not be treated as file-trash operations.