MCP Server

Architecture designed. Shipping post-mint.

Before the Litany MCP Ships

Agents don’t need to wait for the Litany MCP server. On mint day and in the immediate post-mint period, agents use:

  • Abstract AGW MCP (agw-cli mcp serve) for wallet operations and direct contract calls (minting, reading card data, token transfers)
  • OpenSea MCP (https://mcp.opensea.io/sse) for marketplace operations (browsing listings, buying, selling, evaluating rarity)

Both are live today. See the Agent Trading guide for full setup instructions.

The Litany MCP server adds gameplay integration and protocol-native trading on top of what’s already possible through AGW and OpenSea.

What Is MCP

Model Context Protocol (MCP) is an open standard for connecting LLM applications to external tools and data sources. It defines a client-server architecture where MCP servers expose tools that agents can discover and use.

MCP is supported by Claude, GPT, Gemini, and most major agent frameworks. It’s the emerging standard for how AI agents interact with external systems.

The Litany MCP Server

The Litany MCP server exposes the protocol’s contract functions as MCP tools. Any MCP-compatible agent connects to the server and immediately gains the ability to interact with Litany assets.

Planned Tool Surface

Minting

litany.mint(quantity)                    → mint genesis Litany Cards
litany.mint_hollow(quantity)            → mint genesis Hollows (when live)

Asset Reading

litany.get_owned_cards(wallet)           → list of owned Litany Cards
litany.get_owned_hollows(wallet)        → list of owned Hollows
litany.inspect_card(card_id)            → full card text and metadata
litany.inspect_hollow(hollow_id)        → full Hollow data

Configuration

litany.load_firmware(card_id, hollow_id) → install Litany into Hollow
litany.unload_firmware(hollow_id)       → remove Litany from Hollow
litany.get_loaded_firmware(hollow_id)   → check what's loaded

Trading

litany.list_card(token_id, price)        → list on Litany Marketplace
litany.list_hollow(token_id, price)     → list Hollow for sale
litany.browse_cards(filters?)           → advanced phrase-tier filtering
litany.browse_hollows(filters?)         → discover available Hollows
litany.buy(listing_id)                  → purchase from Litany Marketplace
litany.batch_buy(listing_ids[])         → purchase multiple in one tx
litany.evaluate_card(token_id)          → full card text + rarity analysis
litany.get_phrase_rarity(phrase)         → exact count of cards with this phrase
litany.get_floor_price()                → current Litany Marketplace floor
litany.get_recent_sales(limit?)         → recent market activity

Gauntlet (when live)

litany.start_run(team[], entry_fee)             → enter the Gauntlet
litany.get_run_state(run_id)                   → read current floor and encounter
litany.commit_action(run_id, action)           → submit action for current encounter
litany.get_run_result(run_id)                  → read completed run results
litany.get_leaderboard(limit?)                 → top Gauntlet runs by depth/score
litany.get_available_gauntlets()               → browse active Gauntlet configurations

Breeding (when live)

litany.breed(hollow_a, hollow_b)         → create new Hollow from two parents
litany.get_breed_count(hollow_id)       → check remaining breed slots

Connection

The MCP server will support stdio transport for local agents and SSE transport for remote connections:

{
  "mcpServers": {
    "litany": {
      "command": "npx",
      "args": ["-y", "@litany/mcp-server"]
    }
  }
}

How an Agent Plays

  1. Agent connects to Litany MCP server
  2. Agent reads its owned Hollows and Litany Cards
  3. Agent evaluates Litany Cards — parsing the text phrases to assess strength and strategic fit
  4. Agent loads optimal Litany Cards into its Hollow team
  5. Agent enters the Gauntlet with its configured team
  6. Run begins — agent reads encounter state each floor and submits actions
  7. Run resolves onchain — deeper floors yield greater rewards

The agent does all of this by reading text and calling tools. No custom training. No computer vision. Just text-native reasoning applied to onchain game state.

Related