# AgentGraph AI

> AgentGraph AI by Casafari ranks every real estate agency and agent in an area against partner profiles, from listing data, inside Claude, ChatGPT or any MCP assistant. Germany.

MCP server: https://agentgraphai.shiny-bird-2f03.workers.dev/mcp (Streamable HTTP, OAuth 2.1 or an organisation key). Casafari clients reach the same tools through https://mcp.casafari.com/.

## Docs

- [AgentGraph AI overview](https://agentgraphai.shiny-bird-2f03.workers.dev/index.md): what it does, how it works, how to connect, questions and answers.
- [Tool reference](https://casafari-mcp-docs.shiny-bird-2f03.workers.dev/docs/agentgraph): every tool with its parameters, in the Casafari MCP docs.
- [Agent skill](https://agentgraphai.shiny-bird-2f03.workers.dev/.well-known/agent-skills/agentgraph-ai/SKILL.md): how an agent should use these tools.

## Tools

- find_location: Finds German cities, districts and states by name and returns their location ids, kinds and parent areas.
- list_profiles: Lists the partner profiles the screens match against, with each one's exact rules, priority and what the listing data cannot show.
- screen_agencies: Ranks the real estate agencies active in a German area against partner profiles: franchise offices losing momentum and growing independent agencies (solo brokers and small teams of about 2-10).
- screen_agents: Ranks the individual agents named on listings in a German area against partner profiles: top agents in a small area, top performers with a capped territory, agents in traditional offices, and brokers in an early phase.
- agency_profile: One agency's figures, quarter by quarter, its team (named agents on its listings), its places, contacts, and every partner profile it matches, benchmarked against its main city (or the area you give).
- agent_profile: One agent's own figures, quarter by quarter, their agency, career across agencies (where Casafari's directory has it), contacts, and every partner profile they match, benchmarked against their main city (or the area you give).
- partner_brief: A one-page brief on one agency or agent for the person who will approach them: key figures against the area, the partner signals they meet, listing quality (photos, descriptions, sole-agent mandates, price cuts, own website vs portals), team or career, a map of their active listings, two or three talking points, and how they may lawfully be approached in Germany.
- territory_map: One row per district or municipality of a German region (state, district or city): active homes, exits in 12 months, number of brokers, exits and active homes per broker, the share held by franchise networks and by independents, your own network's share, the top networks, a centre point for drawing a map, and an opportunity score.
- save_to_watchlist: Saves agencies or agents (ids from a screen) to your organisation's shared watchlist, each measured in the area you give.
- remove_from_watchlist: Removes entries from your organisation's watchlist by id.
- watchlist: Your organisation's watched agencies and agents, with their area, latest priority, active homes and momentum against the area.
- watchlist_changes: Changes on your watchlist since a date (default: the last 7 days), most important first within each day: priority and profile changes, momentum turning, price cuts, slower exits, falling or growing stock, agents moving agency, going quiet.
- area_benchmark: What normal looks like for brokers in a German area: percentiles of active homes, exits, days on market, price cuts and momentum, how the market splits between independents, franchises, banks and online brokerages, and the biggest networks there.