For AI developers

Connect to the graph over MCP

The graph behind the Feel Layer’s first live register isn’t a dataset you have to request access to. It’s a Model Context Protocol server, running today, that any MCP-capable AI client can call directly.

Server name: io.github.perwinroth/got-cosy-graph
Endpoint: https://gotcosy.com/api/mcp
Tools: find_cosy_hotels, get_hotel_feeling
Source: github.com/perwinroth/got-cosy-mcp

find_cosy_hotels: search by feeling, city, or fitget_hotel_feeling: evidence-backed dimension read for one hotel

Quick connect

In Claude Code, add the server with one command:

claude mcp add --transport http got-cosy-graph https://gotcosy.com/api/mcp

For Claude Desktop or any other MCP client that reads a JSON config, add an entry to yourmcpServers block:

{
  "mcpServers": {
    "got-cosy-graph": {
      "url": "https://gotcosy.com/api/mcp"
    }
  }
}

Every hotel returned carries its evidence: which dimensions cleared the mention gate, and how many reviews back each one. Nothing in the graph is a raw score without a count behind it.

Roadmap

The hotel graph is the first register on MCP, not the last. As product registers publish, they join the same graph: one connection, growing coverage, the same evidence-gated rules on every entry.