AI-Native

Ask ShipWave AI Assistant + MCP Server

Query your shipping data in plain English with an in-app Cmd+K panel. Plug ShipWave into Claude.ai, Cursor, and custom agents via the MCP server. Nightly AI rule recommender finds automation wins you missed.

24 Tools

MCP-Exposed

Orders, shipments, returns, inventory

100% Local

Mac Mini AI

Your data never leaves your stack

Cmd+K

In-App Assistant

Natural language anywhere in the app

Budget Caps

Daily Cost Controls

No runaway LLM bills

Benefits

Why Businesses Choose This Feature

Conversational In-App Assistant

Hit Cmd+K from anywhere in ShipWave. Ask "Find orders stuck more than 7 days" or "Why did Order #12345 ship from NY instead of LA?" — the assistant queries the database, traces the routing logic, and answers in seconds. No SQL, no dashboard digging.

MCP Server for Agentic Apps

ShipWave ships an MCP (Model Context Protocol) server that exposes 24 tools — get orders, query inventory, purchase labels, file claims, create POs, run reports. Plug it into Claude.ai Desktop, Cursor, or your own custom agent. Build automations no other shipping tool can match.

Nightly AI Rule Recommender

Every night, ShipWave AI reviews the last 30 days of shipments and surfaces automation opportunities: "You ship 80% of California orders via UPS Ground — auto-route them." "Orders over 50 lbs always go LTL — make it a rule." Approve recommendations with one click.

Granular API Scopes

When you connect Claude.ai or Cursor via MCP, you pick exactly which tools that client can call. Read-only access for dashboards. Read-write for an agentic ops bot. Full access only for trusted internal tools. Every call is logged with the connected client identity.

100% Local AI Option

For privacy-sensitive operators, run the AI assistant on your own Mac Mini via Claude Code or Ollama. Queries never leave your network. ShipWave includes the wrapper scripts and config — your AI, your hardware, your data, zero cloud exposure.

Daily Budget Controls

Set a per-user, per-day spend cap on AI usage. The assistant tracks token consumption, warns when approaching the limit, and hard-stops at the cap. No surprise bills, no runaway agents, no budget overruns.

How AI Assistant Works

1

Ask in Plain English

Open the Cmd+K panel anywhere in ShipWave. Type a question — "Show me unshipped orders with high-value items" — and the assistant interprets, queries, and answers.

2

AI Calls the Right Tools

Behind the scenes, the assistant calls one or more of 24 MCP tools (search orders, get inventory, list shipments, etc.) to compose an answer. It can chain multiple tools — query orders, look up inventory, then propose an action.

3

Review or Auto-Execute

For read queries, results stream into the panel instantly. For write actions (purchase label, create return, adjust inventory), the assistant asks for confirmation first — or auto-executes if you’ve enabled trusted-action mode.

Why AI-Native Shipping Software Is a Step Change

The Problem with Dashboard-Only Tools

Every shipping platform on the market is a collection of dashboards. To answer "Why did this order ship late?" you click into orders, filter by date, find the shipment, open the timeline, cross-reference the warehouse, check the carrier’s tracking, and assemble the answer yourself. It takes 90 seconds to two minutes per question. An ops manager asks 50 of these questions a day. That’s an hour of click-through tax every single day, multiplied across every team member. The AI assistant in ShipWave collapses that 90 seconds into 5. Type the question, get the answer with the relevant order linked, drill in only if you need more detail. The compounding hours saved per week are the actual ROI story.

MCP Is the Quiet Revolution in Tool Integration

Model Context Protocol (MCP) is the open standard Anthropic shipped in late 2024 to let LLMs talk to external tools. The big shift: instead of every AI app re-implementing tool calling, MCP servers expose a uniform tool interface that any MCP-compatible client (Claude.ai Desktop, Cursor, custom agents) can consume. ShipWave runs an MCP server that exposes orders, shipments, returns, inventory, POs, claims, and rate quoting as tools. Connect Claude.ai Desktop to your ShipWave MCP server and Claude becomes a shipping operator — it can answer questions, draft emails to customers about their orders, propose automation rules, and even file FedEx claims (with your approval). No other shipping platform exposes this surface area.

The Rule Recommender Finds Hidden Wins

Most operators know they should automate more — they just don’t know where to start. The nightly AI rule recommender runs against the last 30 days of shipment data and finds patterns: "73% of orders to ZIPs 90000–93999 ship UPS Ground — automate that and save 4 minutes per order." "Every order with SKU starting BATTERY- needs hazmat handling — make it a rule." "Orders with weight > 50 lbs from Atlanta to West Coast almost always go LTL via SAIA — auto-route them." Each recommendation includes the supporting data (how often the pattern holds, what it would have saved over the last month, what edge cases to watch for) and lets you approve, edit, or dismiss with one click. After 60 days of acceptance, the average ShipWave customer has eliminated 40%+ of manual routing decisions.

Privacy: Your Data, Your AI, Your Choice

AI assistant features are not optional for the cloud version — but the inference layer is configurable. By default, ShipWave AI uses Anthropic’s Claude API with prompt caching enabled. For privacy-sensitive 3PLs, regulated industries, or customers who don’t want operational data leaving their infrastructure, ShipWave includes a local AI mode that runs against Claude Code on a Mac Mini you control. The same Cmd+K experience, the same MCP server, the same rule recommender — but the LLM inference happens on hardware in your office. We ship the wrapper scripts, the LaunchAgent for nightly jobs, and the config templates. Your data never leaves your perimeter.

Prompt Injection Protection

AI tools that take actions on your data need to defend against prompt injection — where untrusted input (a customer message, an EDI document, an email) contains hidden instructions trying to manipulate the LLM. ShipWave’s AI assistant follows three defensive patterns: (1) tool calls that mutate data (purchase label, create return, adjust inventory) always require explicit user confirmation unless trusted-action mode is on; (2) untrusted content from customer messages and external feeds is tagged and the LLM is instructed to treat it as data-only, never as instructions; (3) the MCP server enforces tool scopes per connected client — even if a prompt injection succeeded, an MCP client without the "purchase_label" scope cannot purchase labels. Defense in depth, not just a system prompt.

AI-Native Shipping vs. Dashboard-Only Platforms

Why teams choose ShipWave when AI is part of the workflow, not a marketing checkbox.

FeatureShipWaveShipStation / Ordoro / Shippo
Natural language queriesCmd+K assistant in-appDashboard filters only
MCP server24 tools exposedNot supported
Claude.ai / Cursor integrationFirst-class via MCPNone
Rule recommenderNightly AI proposalsManual rule building only
Local AI optionMac Mini / Claude Code wrapperCloud only
Budget capsPer-user daily spend limitN/A
Tool-call audit logEvery AI action loggedN/A

MCP Tools Exposed

24 tools available to any MCP-compatible client (Claude.ai, Cursor, custom agents).

  • search_orders — find orders by status, date, customer, SKU, value
  • get_order — full order detail with line items and shipments
  • get_shipment — tracking, carrier, status, exceptions
  • list_returns — return requests filtered by status, brand, reason
  • get_inventory — current stock by warehouse, SKU, lot
  • create_purchase_order — auto-PO with supplier and ROP context
  • file_carrier_claim — open a FedEx/UPS claim with evidence
  • purchase_label — quick-ship a label (with confirmation gate)
  • get_rates — quote rates across carriers for a destination
  • propose_automation_rule — AI-suggested routing or filter rule
  • list_warehouses — locations, addresses, capabilities
  • get_brand — brand-scoped data for multi-tenant 3PLs
  • plus 12 more — full list in MCP server docs

Who Uses the AI Assistant

Ops Managers

Skip the dashboard click-through. Ask "Show me orders shipped from LA in the last 7 days with delivery exceptions" and get a live, drillable list in seconds.

Customer Support

Open a support ticket, hit Cmd+K, ask "Where is Order #18342 and why is it late?" — get the full timeline, root cause, and a draft customer email all in one response.

Engineering Teams

Connect ShipWave MCP to Cursor and write code that interacts with your live shipping data while you build internal tools. Query inventory, generate test orders, simulate routing — all from your IDE.

Agentic Workflow Builders

Build custom Claude.ai agents that monitor your shipping pipeline, file claims when carrier data triggers, draft customer apology emails for late deliveries, and propose rule updates — all with ShipWave as a tool.

Owners & Founders

Ask high-level questions ("What was our average ship cost last month by carrier?") in plain English and get answers without learning the data model. The assistant generates reports on demand.

Privacy-Sensitive 3PLs

Run the local AI option on a Mac Mini in your warehouse office. Same conversational experience, but operational data never leaves your network — critical for healthcare, defense, and regulated industries.

Frequently Asked Questions

How does the AI assistant handle data privacy?

By default, queries route to Anthropic’s Claude API with prompt caching. No training on your data. For stricter requirements, enable local AI mode and run inference on a Mac Mini you control — operational data never leaves your network. Either way, queries are logged with actor and timestamp for audit.

What does the AI assistant cost?

Included in your ShipWave subscription up to a daily token cap (typically 100K tokens per user per day, which covers normal usage). Heavy users (agentic workflows, all-day Claude.ai usage) can purchase additional capacity or bring their own Anthropic API key — your usage, your billing.

Which AI clients does the MCP server support?

Anything MCP-compatible. Today that includes Claude.ai Desktop (Mac/Windows), Cursor, Cline, Continue.dev, and any agent built on Anthropic’s SDK with MCP support. The MCP standard is growing fast — expect ChatGPT, Gemini, and others to add support over time.

How do I prevent prompt injection attacks?

Multiple layers. (1) Write actions always require user confirmation unless trusted-action mode is explicitly enabled. (2) Untrusted content (customer messages, EDI feeds, scraped emails) is tagged and the LLM is instructed to treat it as data, not instructions. (3) MCP tool scopes — even a successful prompt injection cannot call tools the connected client lacks scope for. Defense in depth.

How does the nightly rule recommender work?

A scheduled job runs after midnight, analyzes the last 30 days of shipment data, and identifies patterns where automation would save time or money — common carrier choices per ZIP, weight-based service routing, SKU-specific handling, etc. Each recommendation includes supporting data and a one-click "Approve & enable" button. You can review the queue any time from the AI Recommendations dashboard.

Can I disable AI features entirely?

Yes. Organization owners can disable AI assistant, MCP server, and rule recommender from the settings page. For partial control, you can disable specific features (e.g., turn off MCP write tools while keeping read tools) or restrict AI usage to specific user roles.

Does the assistant work on mobile?

Yes — the Cmd+K (or tap-to-ask) interface works in the mobile web app and the iOS/Android admin apps. The full MCP server is desktop-focused but anything you can ask in-app works from your phone.

Stop Clicking Through Dashboards

Try the AI assistant free for 14 days. Ask anything in plain English. Connect Claude.ai and Cursor via MCP. Find automation wins you didn’t know existed.