Plug Claude, Cursor, and ChatGPT into your review data

MCP is an open standard from Anthropic that lets AI clients like Claude call tools on external systems safely. Instead of copy-pasting data into a chat, the AI discovers what tools are available on your server, decides which ones to use for your question, and runs them. Think of it as giving the AI a keyboard in your dashboard, except the keyboard only works inside rules you set.

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: paste the mcpServers config into the client's settings, restart, done.

The REST API is for structured, predictable integrations like automations and pipelines. The AI Hub is for interactive AI use. The AI figures out which tools to call based on your question, the REST API expects you to know the endpoint, the body, and the response shape. They cover different jobs. Many agencies run both.

Only if the token you created explicitly has the reviews.pii permission and the AI asks for include_pii. Both conditions must be true for every single call. By default, private feedback and contact tools return the same data with names, emails, and phones stripped out. That is the default for every new AI token, even ones created by the tenant owner.

Yes. 18 tools cover Google Business Profile posts. An assistant can read each location's settings and posts, draft posts from the business's confirmed facts, schedule them, manage the photo library, pause Autopilot and send posts to the client for approval. A person approves every post, from an email link with no login or in the dashboard.

No. Review replies drafted by the AI wait in the Auto Respond approval queue with status approval_requested. Posts drafted or edited by the AI wait under Posts until a person approves that exact version, in the dashboard or from the email link. No tool approves or publishes.

Yes. The MCP endpoint lives on your white-label domain (for example, https://app.your-agency.com/api/mcp). The server name the AI client sees uses your company name. Your API documentation, audit log, and connect UI all use your branding. EmbedMyReviews is not mentioned anywhere an agency customer can see.

Yes. Each connection you create is its own token with its own scope. You can give Claude Desktop read-only access, create a separate token for your VA with contact PII access, and a third token for your agency's internal automation agent with write access. Revoke any of them independently at any time.

120 calls per minute per token, and 600 calls per minute across the entire workspace. Both limits stack. If you hit either, you get a clean 429 response with a Retry-After header so the AI client can back off and retry. In real use, a Claude conversation usually fires 5 to 15 tool calls per question, so these limits are generous.

Yes. Tool name, redacted parameters, duration, row count, and result status land in a per-tenant audit table. Tokens, emails, and phone numbers in the parameters are redacted before storage so the log itself does not become a PII surface. You can view the log from the AI Hub dashboard, filtered by date range, tool, or status. Rows older than 90 days are pruned automatically.

It is included in the $99 per month EmbedMyReviews plan at no extra cost. You do not pay per tool call, per token, or per AI client. Your only variable cost is whatever your AI provider (Anthropic, OpenAI) charges you for their own usage, which is paid directly to them.

The server implements the 2025-06-18 MCP specification, including the structuredContent tool result shape and spec-compliant JSON-RPC 2.0 error envelopes. Older versions (2025-03-26 and 2024-11-05) are still supported so older mcp-remote builds keep working.

Yes. Agency customers (the businesses the agency serves) can get their own AI Hub connections if their plan grants API access and they have the api.mcp permission. The setup is identical, the config snippet shows the agency's white-label domain, and access is scoped to the customer's own organisations and locations only.

No. Local Search Grid tools are gated on tenant ownership, not on the api.access plan option. A customer with API access still cannot see the LSG tools in tools/list, and any direct attempt to call them returns access_denied. Only your agency-owner tokens unlock the LSG suite (list_scans, get_scan, get_scan_data_points, list_scan_snapshots, get_scan_insights, create_scan, run_scan, schedule_scan).

Yes, and that's exactly the point of the schedule_scan tool. Ask your AI client to enable a monthly schedule on a specific scan with the day-of-week and time-of-day window you care about (for example, Monday to Friday between 9 and 5 in your client's timezone). The platform's cron worker picks up scheduled scans every 15 minutes and triggers a new snapshot whenever the next_run_at falls in the configured window. Combine that with the scan-completed webhook to drop the snapshot straight into Slack, Looker, or your own database.

Impressions (search, maps, desktop, mobile), call clicks, website clicks, direction requests, and search query terms, with monthly impression counts. The metrics endpoint also returns the previous period totals so trend deltas come back in a single call. Profile views and Photo views are no longer available from Google's Performance API, so they're omitted. Access is scoped per token to the locations the caller can already see in the dashboard.