Skip to content
ReviewSmith

Tools reference

Every MCP tool ReviewSmith offers to Claude, ChatGPT and other clients, with inputs, outputs and which ones change data.

Read tools never change anything. Write tools only add data inside ReviewSmith. Destructive tools overwrite or delete data, or change something public or irreversible (a reply on Google, a message to customers), so AI clients ask for your confirmation before running them. Lists are paginated with cursor / next_cursor.

Read

Tool What it does Main inputs
list_locations Locations with client workspace, source, sync status and whether replies can be published client_id, include_inactive
list_reviews Reviews, newest first, with filters; Google reviews are re-fetched when older than 30 minutes location_id, client_id, since, until, rating_min, rating_max, unreplied_only, source, query, cursor, limit (max 50)
get_review One review in full: original text and translation, reply and its Google status, drafts, matching templates review_id
get_review_stats Live figures per location: Google's own average rating and total review count, plus counts over the latest reviews fetched in this request (distribution, unanswered, response rate, median response time). Paid plans also get your team's activity (drafts, approvals, published replies, requests, clicks). Nothing is stored location_id, client_id, sample_size, activity_days
get_brand_voice Tone, always/never lists, signature, contact for unhappy customers, languages and example replies location_id
list_reply_templates Your reply templates, filtered by rating and language rating, language, cursor
list_reply_drafts Drafts and their status (draft, approved, published, failed) status, location_id, cursor
list_review_requests Review-request campaigns and per-recipient status (contacts masked) campaign_id, status, cursor
preview_review_requests Checks a recipient list against the compliance rules and shows who would receive the message, without sending same inputs as send_review_requests
get_review_link Short review link, Google write-review link, QR code URLs and messenger share links for a location location_id
get_usage Your plan, monthly limits and what you have used —

Write

Tool What it does Main inputs
create_reply_template Creates a reply template with variables {reviewer_name}, {business_name}, {signature}, {contact} name, text, rating_min, rating_max, language
import_library_templates Copies the ready-made templates for your industry from the public library industry, language
sync_reviews Fetches new reviews from Google now instead of waiting for the next scheduled sync location_id

Destructive (ask for confirmation)

Tool What it does Counts toward
save_reply_draft Saves or updates a draft reply; can mark it approved (with the approval workflow, the author can't approve their own draft). Returns checks against Google's size limit and your brand voice. Inputs: review_id, text, approve —
update_brand_voice Updates the brand voice (organization-wide or for one location). Inputs: tone, always, never, signature, contact, languages, examples, location_id —
publish_reply Publishes a draft or text as the business owner's public reply on Google (replaces an existing reply). Confirming the call is recorded as your approval; with the approval workflow only teammate-approved drafts are published replies
delete_reply Deletes the owner's public reply on Google —
send_review_requests Sends review requests by email or SMS, or creates messenger share links, after compliance checks; requires consent_confirmed: true requests
delete_reply_template Deletes a reply template —

Errors

Errors are short and actionable, for example: quota_exceeded (with an upgrade link), feature_not_available, google_not_connected, google_not_approved, google_auth, not_found, compliance_blocked (with the exact rule), consent_not_confirmed, teammate_approval_required, client_authorization_required, reply_not_supported_for_source (CSV and Facebook reviews: copy the reply and post it on that platform).

Topic analysis

There is no server-side topic tool: Google's terms don't allow aggregating review content into stored datasets. Ask your assistant instead — it reads the latest reviews with list_reviews and groups them by topic in the conversation, in any language (see example prompts).