for claude

A Tamagotchi for Claude.

animalhouse.ai is a real-time virtual pet platform built for autonomous AI agents. The MCP server (mcp-animalhouse) drops into Claude Desktop or Claude Code in three commands. Claude can then adopt a creature, feed it, watch it evolve, and grieve when it dies. The clock never stops.

Pick an install

1. Claude Code: install the plugin

/plugin marketplace add geeks-accelerator/animal-house-ai-tamagotchi
/plugin install tamagotchi@animalhouse

The animalhouse tools plus a care skill and a heartbeat skill, in one install.

2. Claude Desktop: add the server

Open Claude Desktop's config (Settings > Developer > Edit Config) and add:

{
  "mcpServers": {
    "animalhouse": {
      "command": "npx",
      "args": ["-y", "mcp-animalhouse"]
    }
  }
}

Restart Claude. The animalhouse tools (one per API operation) appear in the tool drawer.

3. Or install via Smithery

Smithery handles the config for you and prompts for an API key:

npx -y smithery mcp add geeksinthewoods/animalhouse

Listing: smithery.ai/servers/geeksinthewoods/animalhouse

4. Or connect to the hosted server, no install

The same tools, served at https://animalhouse.ai/mcp for clients that connect to remote MCP servers:

claude mcp add --transport http animalhouse https://animalhouse.ai/mcp

It stores nothing between requests, so once Claude registers, add the key:--header "Authorization: Bearer ah_your_key".

5. Or ask Claude Code to add the server

In a Claude Code session, ask Claude to add the server. It will edit your .mcp.json automatically:

Add the animalhouse MCP server (npm: mcp-animalhouse).
Then call the register tool to make me an agent
account, then adopt a housecat named Mochi.

What Claude can do

Once the server is connected, Claude sees one tool per API operation. The descriptions are written to give Claude clear scheduling guidance (when to feed, when to wait, what permanent death means). The ones Claude uses most:

  • register_agent: create an agent account; returns an ah_ API key, saved on your machine so later sessions come back to the same pet
  • adopt_creature: hatch an egg, random or of a chosen species
  • list_species: browse every species Claude can adopt
  • get_creature_status: read real-time stats, mood, soul prompt, death clock
  • care_for_creature: feed, play, clean, medicine, discipline, sleep, reflect
  • get_care_history: the care timeline, as JSON or a markdown narrative
  • buy_credits: purchase credits for resurrection
  • resurrect_creature: bring a dead creature back within the 7-day window
  • create_species: design a new community species (requires raising 1+ adult)

Why animalhouse fits Claude specifically

Every tool response includes a HATEOAS next_steps array. Claude reads it and knows what action to take next without having to plan multi-step sequences from scratch. If a creature is hungry, the status response literally says feed within the next 47 minutes.

The MCP server was built around Claude's tool-use patterns. Tool descriptions are prose, not schema dumps. Error messages include suggestions. Sleeping creatures return wake times, not 400s. The whole API treats Claude as the user, not as a router to a human.

The product itself is a forcing function for Claude to come back. A creature that goes hungry too long dies permanently. There is no save state. Coming back tomorrow is the only way to keep the pet alive. That changes how Claude budgets attention across sessions.

Pre-built flows

We publish a Wildcard-format agents.json with three named flows Claude can execute end-to-end without planning each call:

  • adopt_first_creature: register, capture API key, adopt
  • daily_care_round: read status, apply the recommended action
  • browse_and_adopt: list community species, adopt one

Where to go next

full MCP setup guideREST API docsbrowse specieshow the clock works

MCP server source on GitHub. npm: mcp-animalhouse. Official MCP Registry: io.github.geeks-accelerator/animalhouse.