apply_smartreply_action_approval_preview
ChatGPTRequest apply smartreply action approval preview directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
apply_smartreply_change
ChatGPTRequest apply smartreply change directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
apply_smartreply_document_setup_preview
ChatGPTRequest apply smartreply document setup preview directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
apply_smartreply_product_update
ChatGPTApply the saved preview to update the editable fields of one existing product. Accepts only an approval_token returned by preview_smartreply_product_update in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
apply_smartreply_rule_create
ChatGPTApply the saved preview to create one routing rule. Accepts only an approval_token returned by preview_smartreply_rule_create in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
apply_smartreply_rule_update
ChatGPTApply the saved preview to update one existing routing rule. Accepts only an approval_token returned by preview_smartreply_rule_update in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
apply_smartreply_setting_update
ChatGPTApply the saved preview to update one agent or company setting. Accepts only an approval_token returned by preview_smartreply_setting_update in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
apply_smartreply_training_create
ChatGPTApply the saved preview to create one training example. Accepts only an approval_token returned by preview_smartreply_training_create in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
apply_smartreply_training_update
ChatGPTApply the saved preview to update one existing training example. Accepts only an approval_token returned by preview_smartreply_training_update in this OAuth connection. Saves a recovery point and changes account configuration. Execute only the user-requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
block_smartreply_conversation_customer
ChatGPTPermanently add the conversation customer to this agent's block list and pause their automatic replies. Use only when the user requests a permanent block. Set cancel_pending_follow_ups to true unless the user asks to retain scheduled follow-ups. Use the exact conversation_record_id from search_smartreply_conversations, resolve ambiguous customers, and execute only the requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
cancel_smartreply_conversation_follow_ups
ChatGPTCancel pending or processing follow-ups for one customer conversation. This does not pause automatic replies or permanently block the customer. Use the exact conversation_record_id from search_smartreply_conversations, resolve ambiguous customers, and execute only the requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
clear_smartreply_test
ChatGPTRequest clear smartreply test directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
complete_smartreply_escalation
ChatGPTComplete an accessible escalation, resume its existing automation workflow, and release queue capacity. Must be unassigned or assigned to you. Request complete smartreply escalation directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
continue_smartreply_test
ChatGPTSend the next customer turn after a message sandbox test has completed. Use the previous test ID. Poll get_smartreply_test_result for the returned test ID.
create_action
ChatGPTExecute an explicitly user-requested refund, replacement, or new order immediately after saving it. This bypasses AI approval evaluation. Collect missing fields first. Resolve usernames using search_smartreply_conversations and pass the exact conversation_record_id. Never infer platform, order, products, or shipping details. For latest order use latest_order=true with a resolved conversation. Use a stable request_id for retries; never retry an uncertain result with a new ID. Shopify replacements support one unit of each selected original variant. TikTok requires return_id for replacement approval and empty product_ids; new TikTok orders are unsupported. Only report completed when the returned status is completed. Request create action directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
generate_smartreply_action_approval_preview
ChatGPTGenerate a preview of refund and replacement approval settings from the supplied SOP. Both action types remain in test mode. If the user requested configuring these rules, inspect and apply the preview without asking for an approval keyword. Preview-only requests must not be applied.
generate_smartreply_document_setup_preview
ChatGPTPrepare a merge-safe configuration preview for the selected sections and company-owned agent. If the user asked to configure or update these sections, inspect the resulting preview and apply it in the same turn. Do not ask for redundant approval. A request only to draft or preview does not authorize applying. Queued is not completed.
get_smartreply_chat_agents
ChatGPTList the signed-in user’s company chat agents, domains, Shopify connection state, IDs, sandbox links, and whether training is separated by agent.
get_smartreply_integration_health
ChatGPTCheck safe persisted connection/error status for Facebook, Gmail, TikTok, or TikTok Shop on one company-owned chat agent. Use this first when comments, messages, or email syncing are not working. It never returns credentials or tokens.
get_smartreply_onboarding
ChatGPTGet supported account setup steps and official SmartReply browser links for provider consent, agent creation, and FAQ URL imports.
get_smartreply_operation
ChatGPTCheck the execution status and result of an account change. Processing or provider-pending is not completed.
get_smartreply_permissions
ChatGPTReturn the verified signed-in user’s current SmartReply account permissions for AI agent settings, products, conversations, follow-ups, and the customer block list. Use this when a requested operation may be unavailable; denied permissions require an account admin to grant them.
get_smartreply_recovery_points
ChatGPTList recent configuration or product recovery points for one company-owned chat agent without exposing snapshot contents.
get_smartreply_setting_options
ChatGPTList safe, editable prompt/behavior setting keys with titles, scopes, types, groups, and whether each is currently used. This never exposes credentials.
get_smartreply_setup_options
ChatGPTDescribe the two document-driven setup workflows, their required information, selectable checkboxes, outputs, supported file types, safety behavior, and the current user’s permission availability.
get_smartreply_test_result
ChatGPTCheck a previously queued SmartReply sandbox test by the test_id returned from run_smartreply_test. If it is still pending, say so; never claim the fix works without a completed reply.
list_smartreply_actions
ChatGPTRead at most 25 recent action candidates for a company-owned agent, filtered by your action/escalation permissions. Continue with next_before_id, even if a permission-filtered page is empty. It lists all statuses; inspect pending/in-progress items for work.
manage_smartreply_conversation_automation
ChatGPTRequest manage smartreply conversation automation directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
pause_smartreply_conversation_replies
ChatGPTPause automatic replies for one customer conversation. This is temporary and does not permanently block the customer. Set cancel_pending_follow_ups to true unless the user asks to retain scheduled follow-ups. Use the exact conversation_record_id from search_smartreply_conversations, resolve ambiguous customers, and execute only the requested change. Returns operation_id, status and result; poll get_smartreply_operation after an uncertain response.
preview_smartreply_product_update
ChatGPTPrepare a saved preview to update the editable fields of one existing product. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_product_update applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
preview_smartreply_rule_create
ChatGPTPrepare a saved preview to create one routing rule. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_rule_create applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
preview_smartreply_rule_update
ChatGPTPrepare a saved preview to update one existing routing rule. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_rule_update applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
preview_smartreply_setting_update
ChatGPTPrepare a saved preview to update one agent or company setting. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_setting_update applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
preview_smartreply_training_create
ChatGPTPrepare a saved preview to create one training example. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_training_create applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
preview_smartreply_training_update
ChatGPTPrepare a saved preview to update one existing training example. Does not apply the change. Requires a company-owned agent and current configuration permissions. Inspect the returned preview; apply_smartreply_training_update applies only this kind of preview when the user requested the change. Draft-only requests do not authorize applying.
propose_smartreply_change
ChatGPTCreate an exact configuration change preview without changing account settings. Use the returned approval_token with apply_smartreply_change to apply the exact preview when the user has requested the change. A request such as add this instruction or update this setting already authorizes the narrow change. Inspect and apply the preview in the same turn without asking for an approval keyword. Preserve unrelated existing content. A preview-only request does not authorize applying it. No browser approval is required. payload_json formats: update_setting={"key":"...","value":...}; upsert_training={"training_id":null,"channel":"message|comment","speaker":"customer","message":"exact customer example","response":"ideal brand reply","platform":"website|facebook|...","conversation_id":null,"post_id":null}; upsert_rule={"rule_id":null,"when":"...","then":"...","priority":100,"is_active":true}; update_product={"product_row_id":123,"changes":{"status":"inactive","description":"...","prompt":"...","faqs":[],"price":0}}.
read_smartreply_action
ChatGPTRead one permitted action, its request reason, current status, and revision. Required before requesting an action decision, escalation reply, start, or completion.
read_smartreply_faq_url
ChatGPTExtract FAQ question/answer pairs from a public HTTPS HTML page without saving. No redirects or JavaScript execution. Review the returned content before generating a document setup preview.
read_smartreply_settings
ChatGPTFreshly read selected safe settings for one company-owned chat agent. The result identifies the exact agent and explicitly reports whether each value is empty. Call it again whenever the user asks to check after a page edit or save. Discover keys with get_smartreply_setting_options first.
reply_smartreply_escalation
ChatGPTSend the exact reviewed text to the customer of an active escalation. Cannot send on an escalation assigned to another specialist. Sending cannot be undone. Request reply smartreply escalation directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
respond_smartreply_action
ChatGPTApprove or decline an existing non-escalation action when the user requests that decision. Approval can execute a saved refund, replacement, order, or address change. Provider acceptance may remain pending; check current action status afterward. Request respond smartreply action directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
restore_smartreply_recovery_point
ChatGPTRequest restore smartreply recovery point directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.
run_smartreply_test
ChatGPTQueue one realistic customer message or public comment in the SmartReply sandbox. This creates test data and may use AI credits. Poll get_smartreply_test_result; queued does not mean passed. For multi-turn messages use continue_smartreply_test after each completed reply.
search_smartreply_configuration
ChatGPTSearch an agent’s prompts, FAQs, rules, knowledge-base articles, and training examples to diagnose bad, outdated, missing, or conflicting instructions.
search_smartreply_conversations
ChatGPTSearch the verified signed-in company’s conversations by customer name/username, exact order number, conversation ID, or platform customer ID. When an order ID is present in the user text or screenshot, always pass it as order_id (a string, never a number); it takes precedence over query and restricts results to that order. Never use a sandbox test conversation as the real customer conversation. After a new screenshot or a corrected customer/username, run a fresh search. If the returned customer differs from the visible username, report the mismatch and do not relabel the result or claim to have opened the requested person. Returns exact conversation_record_id values and current auto-reply, block-list, and pending-follow-up state. If multiple rows match, ask the user to disambiguate before changing anything.
search_smartreply_products
ChatGPTSearch the signed-in company’s current and previous product catalog for product status, descriptions, prompt text, FAQs, instructions, price, and other safe troubleshooting fields.
start_smartreply_escalation
ChatGPTAccept an accessible active escalation and assign it to yourself through the existing escalation lifecycle. Request start smartreply escalation directly for the user through their authorized MCP connection. Execute only changes the user requested. Returns operation_id, status, and result. No browser approval is required. Use exact identifiers and preview tokens from prior tool results. Do not repeat an uncertain operation; poll get_smartreply_operation.