What is Model Context Protocol (MCP) in plain English?

MCP is an open standard from Anthropic that lets AI clients like Claude call tools on external systems safely. The AI discovers what tools are available, decides which ones to use, and runs them. EmbedMyReviews ships an MCP server out of the box so you can connect Claude Desktop, Cursor, or ChatGPT to your live review data.

Which AI clients work with the AI Hub?

Any MCP-compatible client. Today that includes Claude Desktop, Claude Code, Cursor, VS Code with Copilot Chat, and ChatGPT custom connectors. The setup is the same for all of them.

Can the AI see my customers' private data?

Only if the token you created explicitly has the reviews.pii permission and the AI asks for include_pii. By default, private feedback and contact tools strip names, emails, and phones before the data ever leaves the server.

What are the rate limits?

120 calls per minute per token and 600 calls per minute across the entire workspace. Both limits stack and return a 429 with a Retry-After header if exceeded.

How much does the AI Hub cost?

The AI Hub is included in the $99 per month EmbedMyReviews plan. No per-call fees from us. You only pay your AI provider (Anthropic, OpenAI) for their own usage directly.

Which MCP spec version do you implement?

The server implements the 2025-06-18 MCP specification with the structuredContent result shape and JSON-RPC 2.0 error envelopes. Older spec versions (2025-03-26 and 2024-11-05) are supported for backwards compatibility.

All Features

AI Hub (MCP Server)

Plug Claude, Cursor, and ChatGPT into your review data

The AI Hub exposes your live review data to any Model Context Protocol (MCP) compatible AI client. Ask natural language questions like "summarise my clients' reviews from the last 30 days" or "which locations have the worst response rate", and get real answers from the live database, scoped to the token you created.

Thirteen tools for reviews, organisations, locations, campaigns, contacts (with per-contact activity timelines), metrics, AI insights, auto-respond rules, and more. Read-only by default. PII redacted by default. Every call audit logged. Fully white-label under your own domain.

Start free trial Read setup guide

13 tools, read-only by default PII redacted by default 100% white-label

claude_desktop_config.json

{
  "mcpServers": {
    "my-agency-reviews": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.your-domain.com/api/mcp",
        "--header",
        "Authorization: Bearer ..."
      ]
    }
  }
}

Under a minute to wire up

AI clients that speak Model Context Protocol can now use your reputation data as a working tool, not a copy-paste screenshot.

Model Context Protocol (MCP) is an open standard from Anthropic that lets AI clients safely call tools on external systems. EmbedMyReviews ships an MCP server out of the box. Your AI tool of choice connects over HTTP with a scoped bearer token, discovers the 12 available tools, and starts answering reputation questions from live data. No CSV exports. No brittle screen scraping.

The problem

Your reviews are locked in dashboards

Every agency owner has the same workflow. Open Claude, paste a screenshot of the review dashboard, ask a question, get half an answer. Or export a CSV, drop it into ChatGPT, remember the AI can't tell you anything the file didn't say. Or copy three tabs worth of stats into a Notion doc and give up. Your review data is stuck in the tool, and AI assistants cannot reach it.

The solution

Let the AI run the queries

The AI Hub turns your reputation data into tools any MCP-compatible client can invoke. Claude can pull live review counts by source, group metrics by month, surface 1-star reviews, compare locations, read private feedback, draft responses for you to approve. The AI runs the query against the live database and returns a clean answer, every time, across all of your customers.

Use cases

What agencies actually ask

These are real questions agency owners throw at their AI assistant once the AI Hub is wired up. No keyword search. Plain English in, structured answers out.

Client reporting

"Summarise last month's reviews for Acme Dental. What were the three most common themes?"

The AI pulls the reviews, groups by sentiment, surfaces themes. You paste the answer into the monthly report.

Multi-location analysis

"Which of my client's 14 locations has the lowest response rate? Show me unanswered 3-star reviews there."

Two tool calls, one answer. Uncovers operational gaps without opening a dashboard.

Campaign autopsy

"How did the October Google campaign perform? Compare it to the July one."

The AI pulls the funnel stats, computes deltas, explains what changed. Ready-to-send email to the client.

Sentiment triage

"Pull all 1 and 2 star reviews across my portfolio from the last 7 days. Flag anything that looks like a service failure."

Daily stand-up prep in 20 seconds. The AI reads every new review and tells you what to look at.

Response drafting

"Draft a reply to review 8421 in the voice of the dental clinic owner. Hold for my approval."

Drafts always enter the Auto-Respond approval queue. Nothing ships until a human clicks approve.

Theme tagging

"Group all reviews mentioning wait times, and tag them 'wait-time-issue' so I can filter later."

The AI tags in bulk (up to 50 per call). Tags are visible across the platform, usable in campaigns and filters.

Private feedback review

"Show me the last 20 private feedback submissions. I need to understand what customers are saying privately before posting publicly."

PII is redacted by default. Your approved tokens can opt in to see names and contact info when you need to follow up.

Proposal support

"Which of my clients have no auto-respond rule set up? I want to pitch it as an upsell."

One query turns your account into a list of upsell prospects. Sharp way to find idle revenue.

Team onboarding

"Our new account manager should be able to answer 'how is client X doing?' without logging into the dashboard."

Create them a scoped token, give them Claude, they can read the data they're allowed to see. Nothing more.

Tool catalog

Thirteen tools out of the box

Every tool enforces per-organisation and per-location access checks. Read-only by default. Write tools require a separate opt-in on the token and always route through the human approval queue.

Read (7 tools)

list_reviews

Filter and list reviews across organisations and locations with source, date, rating, sentiment, response status, verified flag, and full-text search. Paginated.

get_review

Fetch a single review by ID with reply, tags, auto-respond history, and language.

list_organizations

List accessible organisations with review count and average rating. Pass organization_id for detail including locations.

list_locations

List accessible locations with per-location review count and average rating. Scope to one organisation.

get_metrics

Aggregate metrics over a date range: totals, avg rating, star distribution, response rate, sentiment. Group by day, week, month, location, organisation, source, rating, or sentiment. Supports period-over-period comparison.

list_campaigns

List review-request campaigns with status, schedule, location, and full funnel (invited, opened, clicked, reviewed, redirected, testimonials submitted, private feedback, unsubscribes, bounced) with conversion and open rates.

get_ai_insights

Retrieve pre-computed AI analysis (themes, sentiment shifts, recommendations) for an organisation or location.

Read with PII access (4 tools)

list_private_feedback

List private feedback submissions (ratings and messages never published publicly). PII redacted by default.

list_contacts

List review-request contacts with subscription state, latest activity, and per-contact engagement counters (invites, opens, clicks, redirects, testimonials, private feedback). PII redacted by default.

get_contact_activity

Full timeline for one contact: every invite, open, click, video play, redirect, testimonial, private feedback, and unsubscribe, with the campaign, channel, and ratings attached.

list_auto_respond_rules

List configured auto-respond rules with rating range, sources, delay, approval requirement, and AI usage.

Write, held for approval (2 tools)

draft_review_response

Draft a response to a review. Always held for agency approval. Never auto-sent. Enters the existing Auto-Respond moderation queue.

tag_reviews

Attach or detach tags on a set of reviews. Useful for clustering after summarisation. Capped at 50 reviews per call.

See every tool with parameters and examples in the API reference

Security model

Built to be safe for AI to touch

AI assistants are still new tools. The AI Hub assumes that, treats every token like a limited guest, and protects your data at every layer.

Scoped by organisation and location

The AI only sees the organisations and locations the token's owner has access to. Try to query an organisation outside that scope and the tool returns a clean access_denied error. No leaks across clients.

PII redacted by default

Private feedback and contact tools strip names, emails, and phone numbers unless the token explicitly holds the reviews.pii permission and the caller asks for include_pii. Two conditions, every call, no accidental PII in AI provider logs.

Write tools never auto-send

AI-drafted review responses always enter the Auto-Respond approval queue with status approval_requested and send_at null. Nothing reaches Google, Facebook, or any review platform until a human clicks approve.

Every call audit-logged

Tool name, parameters (with secrets and PII stripped), duration, rows returned, and result status are recorded for every call. View the audit log from the AI Hub dashboard. Spot misuse, trace a weird answer, keep a clean record.

Rate limited per token and workspace

120 calls per minute per token. 600 calls per minute across the whole workspace. Exceeding either returns a clean 429 with a Retry-After header. One misbehaving client cannot take down the others.

Revocable in one click

Each AI client gets its own named connection. Claude stops behaving? Revoke that connection alone. The others keep working. No "rotate every key" panic.

Under a minute

Three steps to wire it up

If you already have Claude Desktop installed, the whole thing takes under a minute. Cursor and Claude Code work the same way.

Create an AI token

In the dashboard, open Settings then AI Hub and click Create connection. Name it after the client (for example "Claude Desktop"). The token is shown once, copy it right away.

Paste the config

Copy the generated mcpServers block straight into your client config. For Claude Desktop, that's Settings then Developer then Edit Config. Save and restart.

Ask a question

Try "summarise my reviews from the last 30 days" or "which location has the best response rate". The AI discovers the tools, picks what it needs, and answers from live data.

Full setup guide

MCP vs API vs CSV exports

You have three ways to get your data out of the platform. The AI Hub is the right layer when the consumer is an AI client. Here's the honest comparison.

Capability	AI Hub (MCP)	REST API	CSV exports
Natural language questions	Yes	No (structured only)	No
Live data, always fresh	Yes	Yes	Stale the moment you download
Works in Claude, Cursor, ChatGPT	Yes, natively	Needs a custom connector	Paste and pray
PII redaction on by default	Yes	Manual	No
Per-call audit log	Yes	Request logs only	Export events only
Writes require human approval	Always	No, caller controls	N/A
Best for	Live AI assistant work	Custom automations and pipelines	One-off analysis in a spreadsheet

The AI Hub and the REST API cover different jobs. Many agencies use both, the API for scheduled syncs and the AI Hub for interactive work.

White-label

Your domain, your brand, your AI Hub

The MCP endpoint lives on your own white-label domain, for example https://app.your-agency.com/api/mcp. The server name returned to the AI client uses your company name. Nothing in the configuration snippet, audit log, or error messages mentions EmbedMyReviews. Your team and your customers see only your agency.

Custom domain

MCP URL uses your white-label domain with automatic SSL.

Branded server name

Claude Desktop shows "Your Agency MCP Server" in the tool list.

No platform attribution

Your docs, your dashboard, your tokens. EMR never appears.

AI Hub FAQ

Stop pasting screenshots into Claude

The AI Hub is included in every EmbedMyReviews subscription. $99 per month flat. No per-call fees. Wire up your first AI client in under a minute.

Start free trial Read setup guide

14-day free trial. No credit card required.