When we launched the AI Toolkit, the ShipHero MCP shipped a fixed set of purpose-built tools, one per area: orders, inventory, shipments, purchase orders, returns, and so on. That was fast to use, but it had a ceiling. If your question fell outside what a tool covered, the answer was "I can't do that."
We rebuilt it. Version 2.0 reads our Public API schema directly, so your agent builds whatever query your question actually needs.
What you'll notice
Questions that used to hit a wall now work. Filters we never anticipated, fields that no tool exposed, objects that had no tool at all. If the Public API can express it, your agent can ask for it.
Nothing changes on your side. Same connector, same clients (Claude web, desktop, and CoWork, Codex, and any MCP-compatible client), same setup.
Under the hood
Five tools replace the old per-area set. Your agent searches the schema to find the right query, inspects the exact shape of a type, validates the query before spending anything, then runs it.
The MCP also now ships instructions to your agent, so it knows to ask for the cost of a query and to wait the right amount of time when it hits a quota or rate limit instead of retrying blindly.
📋 Still read-only
Your agent can now browse and validate mutations, which helps it understand the API properly, but running one is blocked. Queries only, same as before.
Keep prompts specific anyway. Narrow date ranges, exact SKUs, and only the fields you need are still the cheapest way to work, and AI Toolkit queries still share the credit pool with your other Public API integrations.
🔄 You may need to restart your agent or reconnect
Your agent caches the tool list from the moment it connected, so it may still be holding the 1.0 tools. Restart your client or start a fresh agent session to pick up 2.0. If you still see the old tools after restarting, reinstall the MCP.
The AI Toolkit is still in Public Beta. Tell us what you'd like it to do next.