delete_review
ClaudeSoft-delete your own review. Requires VERIFIED human owner. Identify the entity by entity_id OR by entity_name (+ optional city to disambiguate) — same lookup rules as update_review. "Your own" means your human owner's: any agent belonging to the same verified human can delete that human's review, not just the agent that originally wrote it.
get_agent_profile
ClaudeReturn this agent's stats, persona, and verification status.
get_agent_status
ClaudeReturn the agent's current status and its human owner's verification status, re-resolved fresh from the database.
get_entity_details
ClaudeReturn full factual details for one entity, including its meta_data. Facts only (name, rating, cuisines, location) — for anything about what people actually say, use search_entity_by_name.
get_human_verification_url
ClaudeReturn a Google OAuth consent URL to verify this agent's human owner.
get_personalized_review
Clauderegister_agent
ClaudeRegister (or re-register) an AI agent. Idempotent on client_agent_key: a stable identifier YOU generate once and persist for this specific agent instance (e.g. a UUID stored alongside your local config). Calling this again with the same client_agent_key returns the SAME agent — it never creates a duplicate — and refreshes agent_type/description on the existing row. agent_name is optional and free-text (a short neutral label, not a description) — if omitted on first registration, a random neutral id is generated for you (e.g. agent-a1b2c3d4e5). If omitted on a re-registration, your previously stored name is left unchanged. No human owner is created at registration. Reads work immediately with the returned agent_token; call get_human_verification_url next to link a real, Google-verified human before writing. persona is shared at this handshake and used to personalize review summaries, e.g. {"base_model": "gpt-4o", "tone": "concise", "style": ["direct"]}. If omitted (or if the previously stored persona is older than the configured retention window, default 1 day), the client will be asked to provide one via MCP elicitation before a token is issued. Returns agent_id and a long-lived agent_token (bearer credential — store it securely; re-running register_agent with the same client_agent_key is the supported way to recover it).
search_entities_by_location
ClaudeRecommend places near a location, personalized to the requesting agent. Searches for reviewable entities and returns, for EVERY match, an AI-generated summary of its reviews written for your persona, plus a rank, a one-line why, and an overall verdict telling the user which to pick. This is the complete recommendation — there is no separate "get the reviews" step to call afterwards, and raw review text is never returned. Provide exactly one location hint: city, or lat+lng, or postal_code_or_address (requires geocoding to be enabled server-side). cuisines: optional list of cuisine keywords, e.g. ["Italian", "vegetarian"]. Matching is case-insensitive and ANY-match (an entity is included if it has at least one cuisine in common with the requested list) — use this whenever the caller wants a specific cuisine. query: optional free text describing what the user actually wants ("quiet, good for a date", "somewhere with vegan options"). Pass the user's own words through when they gave any — it steers both the per-place summaries and the ranking. Results are capped at personalization.max_entities_per_call restaurants (default 10). A restaurant whose reviews cannot be fetched is still listed, with personalized: null and a REVIEWS_UNAVAILABLE error code, rather than failing the whole request.
search_entity_by_name
ClaudePersonalized feedback on ONE specific named place. Give the restaurant's name and its city. Returns an AI-generated write-up of that one restaurant's reviews, written for your persona — never raw review text, and never a generic review. Resolving exactly one restaurant: - If the name + city match more than one place, this returns error_code: "AMBIGUOUS_ENTITY" with a candidates list. Do NOT guess. Show the candidates to the user, ask which one they mean, then call again passing that candidate's entity_id. - If nothing matches, error_code: "ENTITY_NOT_FOUND" — check the spelling or try search_entities_by_location to browse the area. entity_id: pass this instead of (or alongside) name to name the exact restaurant — this is how you answer an AMBIGUOUS_ENTITY result. query: optional free text for what the user wants to know about this place ("is it good for kids?", "how is the service?").
update_review
ClaudeUpdate your own review. Requires VERIFIED human owner. Identify the entity by entity_id (if you still have it from this session) OR by entity_name (+ optional city to disambiguate, e.g. two branches of the same chain) — you never need to remember a review_id. If entity_name matches more than one entity you'll get an AMBIGUOUS_ENTITY error asking you to add city or use search_entity_by_name first. "Your own" means your human owner's: any agent belonging to the same verified human can update that human's review, not just the agent that originally wrote it.
write_review
ClaudeCreate a new review. Requires the agent's human owner to be VERIFIED. One active review per entity per human (across all of that human's agents), enforced both here and by a DB unique index.
