# MCP Server The **MCP Server** connects assistants such as **Claude** to **one project at a time** so you can work in plain language with **customers**, **review requests**, **reviews**, **ratings**, **feedback**, **messages**, **locations**, **sources**, **tags**, **charges**, and **share templates**—the same operational data your project already holds for reputation and outreach. MCP (**Model Context Protocol**) is a standard way for assistants to use trusted tools. After you connect through **sign-in** (not by hunting down an API key), the assistant can read and change project records **only for the project tied to the URL you paste**. {% hint style="info" %} Open **Settings → MCP Server** in your project and copy **MCP Server URL** from the box there. That address already includes your project identifier (and your agency’s API host when your account runs under a white-label agency). Use it exactly—do **not** substitute a generic host-only link. {% endhint %} {% hint style="info" %} Running a **white-label agency** and need assistants to manage **clients, team seats, or every business at once**? Use [**Agency MCP Server**](/agencies/mcp-server.md) instead; project MCP stays inside **one** customer business. {% endhint %} *** ## MCP Server page in Settings 1. Select the project you want the assistant to use. 2. Go to **Settings → MCP Server**. You will see: * **MCP Server URL** — copy this full value into your assistant’s connector setup. * **Connected apps** — tiles for tools that already completed sign-in; each shows access level and recency, and you can **Revoke** access here when needed. * **Example prompts** — starter instructions you can paste into chat once the connector is live. *** ## What assistants can help with (project scope) Capabilities mirror what your **project-level automation surface** supports today—they focus on **this business**, not your agency console. ### Customers Look up people in your directory; add or remove records; update notes and contact preferences; attach **tags**; archive or restore; stop or resume outreach where supported; cancel queued-but-not-sent messages; inspect related **charges**, **messages**, or **reviews** for one person when reconciling history. ### Review requests (“asks”) Schedule **review requests** (email or SMS according to your strategy, reminders, eligibility, and sending limits). Integrations do not bypass your throttles or unsubscribe rules. ### Reviews Search and read feedback already collected; adjust moderation flags such as **hidden** or **duplicate** when cleaning data; attach **tags** to reviews; **mark reviews as replied** when you handled the response outside the connector (for example on the original platform); remove reviews when upstream reconciliation requires it; generate **share images** using your saved **share templates**. {% hint style="info" %} Project MCP does **not** publish reply text to Google or other third-party sites on its own. Your assistant can still **summarize**, **prioritize**, or **draft wording** in the chat. To send automated replies through the product’s built-in flows, see [**AI Assistant**](/platform/ai-assistant.md). {% endhint %} ### Messages Review outbound history for reporting-style questions (for example delivery-oriented summaries over recent campaigns). Individual messages can be removed when an integration needs to tidy audit data—**new** review requests are created through **asks**, not by inventing arbitrary message rows. ### Locations List, add, edit, or remove **locations** so attribution stays aligned when assistants help import or reorganize multi-site data. ### Charges Record or delete **spend signals** tied to customers when you sync billing or point-of-sale events into the project. ### Sources and tags Maintain the **sources** catalog your reviews roll up under and the **tags** you use across customers and reviews. {% hint style="info" %} When an assistant sets a preset icon on a source or tag (not a custom uploaded image), use two lowercase words separated by one space, such as **fab fa-google** or **fas fa-star**. Omit the icon when you rely on an uploaded image instead. {% endhint %} ### Share templates List saved layouts for social-style review graphics or remove templates that are no longer needed. *** ## Connect Claude (or another MCP assistant) Menus vary by product, but the flow is consistent: 1. Open the assistant’s **Settings**. 2. Open **Connectors**, **Integrations**, or the equivalent. 3. Choose **Add custom connector** (or similar). 4. Paste the **MCP Server URL** from **Settings → MCP Server**—the entire string from the copy box. 5. Save. The assistant opens a **sign-in** window linked to your account. 6. Sign in if prompted. On **Authorize**, pick the **project** when asked (if your connector URL already scopes one project, that choice may be fixed for you), review **read / write / delete** style permissions, and approve. When the connector shows as connected, try one of the **Example prompts** from your MCP Server page or any of the suggestions below. {% hint style="success" %} To move the assistant to a **different** project, remove the connector and add it again using that project’s **MCP Server URL**, then authorize again. {% endhint %} *** ## Example prompts aligned with the product These match the spirit of the built-in **Example prompts** block and stay within what project tooling can actually drive: 1. **Weekly review digest** — Ask for a concise summary of reviews from the last seven days with counts, averages, breakdown by **source** and **location**, recurring praise, and recurring concerns. 2. **Send review requests safely** — Ask the assistant to **list** customers who match your criteria (for example recent sign-ups who have never been asked) and **wait for your confirmation** before scheduling sends with reminders. 3. **Add someone and ask** — Provide contact details for a new customer, then have the assistant create the record and schedule the first review request with your preferred reminder count. 4. **Who still needs an ask** — Request a capped list (for example up to 50) of customers never asked, newest first, skipping unsubscribed contacts, with name, email, and signup date. 5. **Location comparison** — Ask for a ninety-day **leaderboard by location** using review volume, average rating, share of five-star scores, and a simple trend versus the prior ninety days. 6. **Outreach health** — Ask for counts and rates across recent email sends (totals, delivery, opens, clicks, failures) so you can spot anomalies over roughly the last two weeks. Your assistant still reasons in natural language—it chains the underlying lookups and updates for you. {% hint style="info" %} Assistants often pause for **confirmation** before bulk sends, large deletes, or anything similarly impactful. Use that moment to double-check filters and counts. {% endhint %} *** ## Permissions Authorization typically exposes three bands of capability (the assistant only sees actions allowed by what you approve): * **Read** — inspect customers, reviews, messages, locations, charges, sources, tags, and share templates. * **Write** — create or update customers and locations; schedule review requests; record charges; maintain sources and tags; adjust review flags, tags, **mark-as-replied**, or share-image jobs as supported. * **Delete** — remove customers, reviews, messages, locations, charges, sources, tags, or templates when those workflows are permitted. Everything stays inside the **one project** you authorized. Other projects, billing screens, and [**agency-wide**](/agencies/mcp-server.md) controls remain out of scope unless you add a separate connector. {% hint style="warning" %} Treat MCP access like handing someone your project keyboard—prefer assistants you trust, and review destructive actions (deletes, bulk outreach, broad flag changes) before approving them. {% endhint %} *** ## Other MCP-compatible assistants Any client that supports custom MCP connectors can use the same **copied URL** and the same **sign-in** flow—not only Claude. *** ## Disconnect or revoke **Inside the assistant.** Remove the connector from its settings list; access stops immediately. **Inside More Good Reviews.** On **Settings → MCP Server**, use **Revoke** under **Connected apps** for fine-grained cleanup, or revoke tokens from your account security flows if you suspect misuse. Disconnecting does **not** erase reviews, customers, or ratings—it only removes the assistant’s path back into the project. Reconnect any time by repeating authorization with the same **MCP Server URL**. *** ## Tips {% hint style="success" %} Name the product and project in your first message (“In More Good Reviews for *Bright Dental*, …”) so the model reliably chooses this connector when several are installed. {% endhint %} {% hint style="info" %} Save recurring prompts—weekly digests, monthly “never asked” sweeps, outreach audits—as reusable snippets inside your assistant. {% endhint %} {% hint style="warning" %} Assistants may hit practical limits when reading huge histories in one pass. Narrow time ranges (“last 30 days”) or row counts (“latest 100 reviews”) when results truncate unexpectedly. {% endhint %} *** ## Related documentation * [Agency MCP Server](/agencies/mcp-server.md) — assistants scoped to **agency operations** across many client businesses. * [AI Assistant](/platform/ai-assistant.md) — built-in flows for AI-authored replies and copy inside the product. * [API Reference](/platform/api-reference.md) — developer-oriented detail alongside MCP. * [Integrations](/platform/integrations.md) — Stripe, HubSpot, Slack, Google, and other first-party connectors. * [Webhooks](/platform/projects/webhooks.md) — event notifications into your own systems. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.moregoodreviews.com/platform/mcp-server.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.