Run your App Storefrom your IDE

The Model Context Protocol is the standard that lets AI coding agents call external tools. An MCP server for App Store Connect turns store operations (set a price, edit metadata, attach a build, submit for review) into tools your agent can call mid-conversation, which means release chores become sentences in the same session where you wrote the code.

The open-source options are local processes wrapping the raw API: you clone a repo, hand it your .p8 key, and get thin endpoints with no safety layer. MetaRun's server is the other kind: hosted, credential-secured, and built from the same 187 production tools that power its dashboard and AI operator, preview gates included.

The last hour of every release is the same: bump the version, write What's New, translate it, check the screenshots, attach the build, submit. With the MCP server connected, that is a conversation: your agent drafts What's New from the changelog it just helped you write, stages it across locales, attaches the processed build, and walks the submission, previewing each step for your approval.

It also compounds with your own automation: agents can read your tracked ASO keywords before writing metadata, check territory pricing before a launch, or pull review sentiment while triaging issues, because reads are free-flowing and only writes are gated.

The obvious objection is the right one: an AI agent with store credentials is a hazard. The design answer is that the agent never holds credentials and never holds unilateral write power. Authentication is a revocable personal access token (stored as a hash, shown once). Credentials stay server-side. Every mutation requires the preview's confirm token, drift on the store invalidates it, and destructive submissions require explicit acknowledgment flags.

Everything an agent applies lands in the same change history as manual and chat edits, attributed as MCP, and is revertible in one tap. You get the leverage of agentic operation with a paper trail and an undo.