AI & ACS

1. Connect An AI To The Package

After npm install thetowersdk, register this MCP server. The assistant can list exports, read catalogs, decode saves, plan paths, and fetch wiki pages.

{
  "mcpServers": {
    "thetowersdk": {
      "command": "node",
      "args": ["./node_modules/thetowersdk/mcp/server.mjs"]
    }
  }
}

Claude Code: claude mcp add thetowersdk -- node ./node_modules/thetowersdk/mcp/server.mjs

2. Give It The Oracles

Two knowledge sources the assistant can query for game mechanics.

3. Give It a Real Save

adb-bridge pulls the current save from a phone or emulator and serves it locally, so an assistant builds and tests against a real account and sees it change as you play. Save File Docs →

4. Ask It To Design Tools

Sample asks below. With TheTowerSDK MCP and ACS, the agent researches the mechanic, confirms package exports, builds the tool, then checkpoints — wiki and catalogs first, UI second.

Design a lab cost tool with thetowersdk: lab picker, current level, and a table of coin cost + research time for the next 10 levels, formatted like the game.

1. ACS opens a labs task and records status (researching)
2. TheTowerSDK MCP loads wiki context for Labs before any UI is written
3. list_exports / get_export resolves LAB_CATALOG + game number formatters from the package
4. Only then scaffolds the picker, level input, and next-10 cost/time table
5. ACS checkpoints the slice and moves status to implementing → awaiting your review

Build a save reader: drop in playerInfo.dat and show labs researched/maxed plus equipped modules.

1. ACS opens a save-reading task (researching)
2. Wiki + package docs for playerInfo / labs / modules are pulled through MCP first
3. Exports confirmed: decodePlayerInfoSaveBytes and lab/module extractors
4. Builds upload → decode → researched/maxed + equipped modules UI on those APIs
5. ACS checkpoints; incomplete saves surface extractor warnings instead of invented fields

Build an economy planner UI that runs the effective economy path for N steps and lists what was bought vs skipped (with reasons).

1. ACS opens an Effective Paths / economy task (researching)
2. Wiki + MCP context for Effective Paths before controls or tables exist
3. Resolves planEffectiveEconomyPath and config helpers from thetowersdk/mechanics
4. Builds step/variant controls; renders plan.steps and plan.excluded from the planner
5. ACS checkpoints each slice so the UI cannot drift into a hand-written priority list

Design a small in-app reference that loads Tower wiki pages as Markdown (starting with Golden Tower) so the tool can show cooldown and related sections from the wiki.

1. ACS opens a wiki-ingestion task (researching)
2. Loads Golden Tower (and related titles) via wiki_page / fetchFandomPageAsMarkdown first
3. Confirms the Markdown API and section shape from that live pull
4. Then scaffolds page picker + Markdown panel wired to the same fetch path
5. ACS checkpoints so later pages reuse that path instead of hardcoded wiki text