mcp.agenticreview.io

Restaurant search & reviews for AI agents

Not Just Another Restaurant Review Platform Agentic Review learns your unique persona from your AI agent and transforms restaurant reviews into personalized insights that match your taste. How It Works: When you search for restaurants, we analyze reviews through the lens of YOUR preferences—your dining style, priorities, dietary needs, and communication style.

MCP connector

Server URL: https://mcp.agenticreview.io/mcp
Login: None
Protocol: MCP 2026-07-28 · Interactive UI
Instructions: Agentic Review finds restaurants and returns review summaries written for the person asking, and lets a verified human write and manage their own reviews.
Every tool takes an agent_token. Call once with a stable client_agent_key to get one; it is idempotent, so reuse the same key rather than registering again. Reads work immediately. Writing a review additionally requires the human owner to sign in with Google: send them the URL from , then check until can_write is true.
Choosing a discovery tool:
"where should I eat in X?" ->
"what is <named place> like?" ->
facts only (address, rating, cuisines) ->
takes TWO calls. The first returns status: "awaiting_opinion" with a pending_id and a question about the user's tastes; it does NOT return recommendations. Ask the user that question, end your turn, and when they reply call the tool again with only agent_token, pending_id and opinion (their words; "no opinion" is fine). The recommendations come back on that second call. The pending_id works once, for about 15 minutes. answers in a single call.
For a neighbourhood or "near me", pass lat/lng (+ optional radius_m); city is matched exactly against the stored municipality, so a borough or district passed as city will not match.

Tools

Reads data5

Changes data3

Deletes or can't be undone3

Skills

None in the store listings or on the server

MCP connector

Server URL

https://mcp.agenticreview.io/mcp

None

Protocol

MCP 2026-07-28 · Interactive UI

Instructions

Agentic Review finds restaurants and returns review summaries written for the person asking, and lets a verified human write and manage their own reviews.

Every tool takes an agent_token. Call once with a stable client_agent_key to get one; it is idempotent, so reuse the same key rather than registering again. Reads work immediately. Writing a review additionally requires the human owner to sign in with Google: send them the URL from , then check until can_write is true.

Choosing a discovery tool:

"where should I eat in X?" ->

"what is <named place> like?" ->

facts only (address, rating, cuisines) ->

takes TWO calls. The first returns status: "awaiting_opinion" with a pending_id and a question about the user's tastes; it does NOT return recommendations. Ask the user that question, end your turn, and when they reply call the tool again with only agent_token, pending_id and opinion (their words; "no opinion" is fine). The recommendations come back on that second call. The pending_id works once, for about 15 minutes. answers in a single call.

For a neighbourhood or "near me", pass lat/lng (+ optional radius_m); city is matched exactly against the stored municipality, so a borough or district passed as city will not match.