Use case · Reputation & compliance
Connect Claude, ChatGPT or Cursor to Grades.com through the MCP server or the REST API. Your agent reads every new review, the drafted replies, the reviews that may break a site's rules and what AI assistants say, and tells you what needs a person today. Nothing posts before someone says yes.
review_list7 since 6 pm: Google 5, Yelp 1, Facebook 1.agent_actions4 drafts in your voice; the 2-star Yelp review is held for you.review_handlingA Google review describes a shop in another town. Report steps attached.Today: approve four drafts, call one customer, decide on one report
Start the day with a list, not a dashboardYour agent reads every site you connect and answers in a sentence: what came in, what is waiting, what needs a call.
The rules are built inThe same ask for every customer, each site's own rules checked on every review, and a record that is never thrown away.
A person keeps the last wordReplies, reports and requests wait for a yes, from you or the approvers you name.
What your agent can do here
Each one is a sentence typed into Claude, ChatGPT or Cursor. The agent calls the Grades.com tools named under it and answers from the client's data.
"What came in overnight across all our sites, what needs a reply today, and does anything break a site's rules?" The agent reads new reviews with review_list, the Review Agent's drafts and its check of each review against the site's published rules with agent_actions, and the open Review Disputes cases with recovery_cases. For a group, locations_agent_actions covers every location.
"Rewrite the draft for the 2-star Yelp review in a calmer tone and hold it for Dana." draft_reply writes in the business's voice, agent_approvers names who may approve, and agent_approve runs only after that person's yes. Google replies post directly; other sites get the text and a link.
"Which reviews from the last 30 days appear to break the rules of the site they are on? Draft each report in that site's words." agent_actions returns each review's policy findings, review_handling gives the site's own reporting steps, and once you file, flag_review and update_flag track the outcome in Review Disputes. The site decides, and an honest bad review is not a violation.
"Ask everyone we served this week for a review, the same message for each." review_request_campaigns and review_page_get show what each customer will get, review_request_settings keeps anyone from being asked twice too soon, and review_request_send goes out only after your yes. No gating and no incentives.
"What do ChatGPT, Perplexity, Gemini and Claude say about us this week, and did anything change?" ai_perception returns the stored answers, the sites each cites and every quoted rating checked against your reviews; alerts_list shows what changed. ai_probe_run asks again, once a day.
Connect it once
Add the MCP server to Claude, ChatGPT or Cursor and sign in, or give your own tools a scoped API key. Webhooks tell them when a review lands or a draft is waiting. No AI client? Ask Grades, the assistant in your dashboard, answers the same questions.
MCP server https://mcp.grades.com/mcp
REST API https://app.grades.com/api/v1
Scopes reviews:read, companies:read; add reviews:write only for the person who approves
ChatGPT custom connectors need a Business, Enterprise or Edu plan
Compliance
The FTC rule on consumer reviews: no fake reviews, no reward that depends on what a review says, and no hiding bad reviews while showing the rest as all of them. Review Generation asks every customer the same way and shows every site to every visitor.
The rules of reviews →Google, Yelp, Facebook and the rest each publish what a review may contain and how a business may ask for one. The Review Agent checks every new review against the rules of the site it is on, and Yelp never gets a button on your review page.
Grades.com reads the sites' guideline, reply and reporting pages every week and posts the changes that matter, each dated, with links to the page before and after.
Review policy changes →Review Archive keeps every review we collect with its rating, text, date and your reply, including ones a site takes down or a reviewer rewrites. Nothing is deleted; each edit is kept as a version.
What we never promise
We report only reviews that appear to break a site's rules. The draft names the rule and quotes the evidence; you file it, and the site decides. Most negative reviews stay up, and a reply is often the better answer.
Review Disputes →ChatGPT, Gemini, Perplexity and Claude asked what a customer would ask, with the sites each one cites and the ratings it quotes checked against your reviews.
AI Reputation →Approved replies post on Google for you. For Yelp and the other sites you get the text and a link to post it yourself.
Review Agent →What we can and cannot do. Nobody controls what an assistant says, and we never promise to change an answer. Nobody can promise a review comes down either. We show what is said, where, and the day it changes.
What the account owner gets
Each one runs per location, in the dashboard and through the MCP server and the API.
| Deliverable | Product | MCP tools | Plans |
|---|---|---|---|
| A reply draft for every review, approved by the person you name | Review Agent | agent_actions, agent_approve, draft_reply | 10 drafts a month on Starter, 50 per location on Pro, no cap on Premium |
| Reports for reviews that appear to break a site's rules, drafted and tracked | Review Disputes | review_handling, flag_review, update_flag | Drafts ready to file on Pro and Premium |
| A review page and QR code per location, the same ask for everyone | Review Generation | review_page_get, review_request_send, review_request_stats | Every plan; request campaigns on Premium |
| Every review kept, including ones a site takes down, with each version | Review Archive | review_list, review_history, reviews_export_csv | Every plan |
| Dated answers from four assistants, with cited sites and rating checks | AI Reputation | ai_perception, ai_probe_run, alerts_list | Every plan: monthly on Starter, weekly on Pro, daily with alerts on Premium |
| The same questions, answered in the dashboard with no setup | Ask Grades | None to set up | 50 questions per location a month on Starter, 300 on Pro, 2,000 on Premium |
Nobody can promise that, and we do not. If a review appears to break the rules of the site it is on, we draft the report with the rule named and the evidence quoted, you file it, and the site decides. We record whether it is still up after 30 days.
Yes, and that is the point: the same request to everyone, with no reward and no question about how happy they are first. Asking only happy customers is review gating, which Google forbids.
Only if you set it up that way, on Premium, for routine reviews. By default every draft waits for a yes, hard reviews always wait, and every write is logged with who made it.
It stays in your Review Archive with the date it disappeared and every earlier version, so you can show what was said and when. If it comes back, it is restored by itself.
No. We build the rules we know into the product and link the sources; your own counsel decides what applies, especially in regulated professions.
Product names and logos are trademarks of their owners; no endorsement implied.