add_files_to_directory
ChatGPTAdd existing files to a folder (directory). Ids this key cannot see are reported under rejected.
approve_file_type
ChatGPTApprove one file type (blueprint) against schemaId.
create_directory
ChatGPTCreate a new folder (directory) in the workspace.
create_file_preview
ChatGPTCreate a short-lived preview URL for a stored file, given its s3Path (returned by request_file_upload).
create_upload_link
ChatGPTCreate a link where the user uploads one file to fileAI from their own browser. Use this when your requests cannot reach S3: agents in a desktop or web chat app (Claude.ai, Claude Desktop, ChatGPT), whose network sandbox blocks storage hosts, so a presigned URL PUT fails. Nothing needs to be encoded, whatever the file size. Coding agents whose shell commands run on the user's computer use request_file_upload first, then upload_file, and this link only when neither works. If the user already gave a public https link to the file, use upload_file with sourceUrl instead. Show uploadLink to the user as a clickable link and ask them to open it, choose the file and tell you once it has uploaded. The link takes one file and expires after 30 minutes. Then end your turn, because the user may not see the link until you do. After they reply, call wait_for_upload_link with the linkId and waitSeconds 50.
delete_directory
ChatGPTDelete a folder (directory) by id. Files inside are not deleted. Requires confirm: true.
delete_file_types
ChatGPTDelete file types (blueprints) by id. Requires confirm: true.
delete_files
ChatGPTDelete files by id. Soft-deletes by default (recoverable); set isHardDelete to true to delete permanently. Requires confirm: true.
delete_forensics_run
ChatGPTDelete a forensics run by its analysis_id (run id). Requires confirm: true.
export_schemas
ChatGPTExport one or more schemas (blueprints) by id. Import is not available via this tool.
get_file_forensics_result
ChatGPTGet the status and forensic report of an analysis started with run_file_forensics. While it is queued or running, only the status is returned and nothing is charged. Once complete, the report and the charge summary are returned; asking again does not rerun or recharge. Returns a compact per-check summary by default; for the full report with every check's file_checks detail, call again with view: "full" and confirmFullView: true.
get_file_ocr
ChatGPTGet the OCR text of a file.
get_file_type_schemas
ChatGPTGet the schemas that belong to a file type (blueprint).
get_file_values
ChatGPTGet the extracted values of a file. Returns a compact view by default: each field title mapped to its value, and the form value id that update_file_values needs. A table is { columns, rows }: the column names once, then each row as a list of values in that order (a row’s position is its rowIndex for update_file_values). Wait for processing first (wait_for_upload until its status is processed): values only exist once extraction has finished. A 409 with code NOT_READY means extraction is still running; wait retryAfter seconds and try again. A 409 with code EXTRACTION_FAILED means extraction finished without values; retrying the read will not help. The full view (view: "full") repeats every field definition and adds citations and validation reasoning, often tens of thousands of tokens. Never request it on your own: tell the user how many tokens it will use (the compact response estimates it in fullView) and ask for their explicit permission; only then call again with view: "full" and confirmFullView: true.
get_project_settings
ChatGPTGet the project settings.
get_subscription
ChatGPTGet the workspace subscription.
get_workflow
ChatGPTGet a workflow by id.
get_workflow_result_groups
ChatGPTGet workflow results grouped by folder or subfolder ids.
list_directories
ChatGPTList folders in the workspace, optionally by name search.
list_file_types
ChatGPTList the file types (blueprints) in the workspace.
list_files
ChatGPTList files in the workspace, optionally filtered by ids, directory or update/create time.
list_projects
ChatGPTList the fileAI projects you can work in, with their workspace. Pass a project id as project_id to the other tools.
list_schemas
ChatGPTList AI schemas, optionally filtered by schema or upload ids.
move_files_to_directory
ChatGPTMove files from one folder (directory) into targetDirectoryId. Nothing is deleted.
remove_files_from_directory
ChatGPTDetach files from a folder (directory). The files stay in the workspace and in any other folders; nothing is deleted.
rename_file_type
ChatGPTGive one file type (blueprint) a new name. To delete file types, use delete_file_types.
request_file_upload
ChatGPTUpload one file to fileAI using a presigned URL. Use this tool only if you can run shell commands (such as curl) or otherwise send an HTTP PUT request yourself; if you cannot, use upload_file instead. Returns presignedUploadURL, uploadId and s3Path. PUT the raw file bytes to presignedUploadURL within 5 minutes, sending no x-api-key (the URL is already signed), e.g. curl -X PUT --upload-file ./invoice.pdf "<presignedUploadURL>". If the PUT is blocked by an egress proxy (CONNECT ... 403), allowlist the host from presignedUploadURL (see onUploadBlocked in the response) at claude.ai Settings → Capabilities, or use upload_file for files within the cap. Then call wait_for_upload with the uploadId to wait until processing finishes.
rerun_form_filling
ChatGPTRerun form filling (extraction) for a file, optionally limited to specific field paths.
retry_forensics
ChatGPTRetry a forensics analysis by its analysis_id (run id).
run_file_forensics
ChatGPTStart a forensic integrity analysis of a file already stored in fileAI. Each run is charged once, only when it completes; the quoted price is returned. Returns an analysis_id: call get_file_forensics_result with it for the status and report. Pass the same idempotency_key when retrying a request so it is not run or charged twice.
update_file_export_status
ChatGPTSet a file's export status as reported by an integrated system.
update_file_values
ChatGPTUpdate a file's extracted values (scalar fields and/or table cells). Provide the schemaValueId from get_file_values.
upload_chat_file
ChatGPTUpload a file the user attached in this chat (ChatGPT) to fileAI. Pass the attached file as file; the chat app fills in its link. Use upload_file or create_upload_link when the file is not attached in the chat. Returns uploadId; call wait_for_upload with it to wait until processing finishes.
upload_file
ChatGPTUpload one file to fileAI in a single call, with no shell needed. Use this when you cannot run shell commands (such as curl) or send an HTTP PUT request yourself; otherwise use request_file_upload. Provide exactly one of: sourceUrl, a public https link that fileAI downloads (up to 50 MB); or contentBase64, the file bytes base64-encoded (up to 10 MB). Pass fileName whenever you know it; the file type is detected when fileType is omitted. Returns uploadId; call wait_for_upload with it to wait until processing finishes.
wait_for_upload
ChatGPTWait until an upload finishes processing, then return its status. Checks every 5 seconds for up to timeoutSeconds (default and maximum 50) and returns as soon as the status is processed, action_required or failed; timeoutSeconds 0 checks once without waiting. If processing is still running when the time is up, done is false: call wait_for_upload again. The top-level status is the overall answer: processed (every document is done; read the values with get_file_values for each fileId), action_required, failed, processing or uploading (the file bytes have not arrived yet). detail "summary" (the default) returns a short status summary; detail "full" returns the complete upload record with step-by-step progress, error reasons and the contents of a ZIP. The statuses list in the full record shows individual steps for progress only; it does not include extraction, so do not use it to decide when processing is done.
wait_for_upload_link
ChatGPTWait for the user to upload a file through a link from create_upload_link. Checks every 3 seconds for up to waitSeconds (default and maximum 50) and returns as soon as the file has arrived; waitSeconds 0 checks once. status is waiting (not uploaded yet), uploading (the file is on its way), uploaded (with the uploadId to continue with) or expired (create a new link). If it is still waiting after a second call, ask the user to tell you once they have uploaded the file, then call it again.
whoami
ChatGPTShow which fileAI account this connection uses: the API key name, the project and workspace (ids and names), the plan and the environment. Use it when the user cannot find a file in the fileAI app (it may be in another project, workspace or environment), or to confirm where uploads will go before uploading.