How your agent works with Craft Anything
Everything Craft Anything can do is a command in its own command registry: the same commands as its menus and shortcuts. Your agent finds them and runs them; it does not need a list from us in advance.
- Upload your file with
create_upload, then open it withanything_open_document. - Read it with
anything_read_document. - Find the commands you need with
anything_list_commands, searching with a few words, and read their parameters. - Run them with
anything_run_command. If a command reports an error, adjust the parameters and run it again. - Save or export with a command, download the result with
anything_get_file, thenanything_close_document.
Tools
| Tool | Purpose |
|---|---|
anything_open_document | Start a private session: open an uploaded file, or a new blank document. Returns document_id and revision. |
anything_list_commands | List Craft Anything's commands, read live from the editor running in this session: every command id with its label and parameter documentation. Search with a few words or page through all of them. Free. |
anything_read_document | Read the current document in this session. |
anything_run_command | Run a command found with list_commands, with its documented parameters. Include the current revision and a unique request_id; a repeated request_id returns the first result instead of running twice. |
anything_quote | See what a request will cost before running it. |
anything_stage_file | Copy another uploaded file (an image, font, media or resource) into the session for a command that needs it. |
anything_get_file | Get a file the session produced (an export or saved copy) as a private download. |
anything_close_document | Close the session after downloading what you need. This ends its undo history. |
Long operations return a job_id: poll get_job until it finishes. Commands that need a desktop window, a printer, a local device or a local file path cannot run in a server session.
The website Agent
The website Agent explains the open viewer’s file identification and triage evidence. It is read-only and can return an evidence report. Native MCP supports deeper uploaded-file analysis; it does not control the browser. Hosted sessions use restricted mode, so host-machine, device, network and website capture actions are unavailable.
Resources and saved copies
Projects that reference media, fonts or other files need those files too: upload them and add each with anything_stage_file. Sessions can end on close, idle expiry or service restart, so download your results. MCP files are temporarily stored on EasyRunCraft servers for processing. They expire after 2 hours without activity on that file. Processing, file quotes, result retrieval and downloads renew the idle deadline. Ordinary chat and capability discovery do not. Active processing and transfers are protected from cleanup. Expired files cannot be accessed and are deleted automatically while the service is running; cleanup resumes on restart after an outage. Keep the downloaded edited results: local copies are not deleted by server expiry. Account and billing records are retained separately. You can ask your agent to delete a remote file version earlier.
Credits
Each successful MCP atomic operation costs 1 credit, including file inspection, editing, audio conversion, video export and animation rendering. One complete task may need several atomic operations. Credits are reserved before execution and charged only on success. Failed or cancelled tasks release the reservation. Uploads, quotes, capability checks, status polling and downloads have no separate charge. MCP credits pay for file processing, not AI model tokens. Your external agent’s model usage is billed by its provider.