ASO with Claude Code and AI agents: the board as a tool

You already direct an AI agent to write code, and you would like it to handle keyword research next to the code instead of in a separate dashboard.

• Copy and paste between tools. Numbers live in one app, decisions in another.
• Agents that guess. Without data an agent invents plausible keywords.
• Trust. You want the agent to read and add keywords, not to ship metadata on its own.
• Setup friction. Many tools need an account and an API key before an agent can use them.

1. Install and activate LocalASO, then add your app.
2. Connect your client to the MCP server. The installed app runs the server itself, with a short command for Claude Code and a config file entry for Cursor or Codex. The Claude Code walkthrough shows it in full.
3. Ask the agent to read the board with board_get, propose keywords from your code or README, add them with board_add_keywords and check them with keywords_check.
4. Ask for competitors with competitors_get and the saved listing with listing_get, so the agent's text stays close to what is live.
5. Review the board in the app. The verdicts and ranks are the same numbers the agent saw.
6. Use Optimize for the final text, and upload through the App Store Connect key only as a draft that you submit yourself.

For the background, read what an ASO MCP server is.

LocalASO uses an AI you bring (Claude Code, an Anthropic API key, Ollama, Codex or Gemini CLI) to adapt keywords and write text. Your own agent, separately, calls LocalASO through MCP. Configure each; they do not need to be the same provider.

• Not a cloud API. The server is stdio on your Mac and opens the same local database as the app.
• No data leaves via MCP except the public Apple search requests that rank checks make.
• The numbers are estimates from rating counts, and an agent's recommendation inherits that limit.
• A license is still needed for the tools to work.

Give the agent a narrow job and a stopping rule. For example: read the board, list the Push keywords with their ranks, propose five new candidates from the README, add them, check them in two countries, and report which ones came back Dead or Hard. Keep the human decision for the final text. An agent can be wrong about what a phrase means in another language, so ask it to flag every phrase it is unsure about and let the app's rank check confirm the rest.

Add one app, connect one client, and ask for ten keywords you can check. Download or read pricing. Related: privacy-focused teams and budget-conscious developers.