Key Takeaway
JASCI is the first AI-native warehouse management platform to ship a production Model Context Protocol (MCP) server. Starting August 2026, customers can point Claude Desktop, ChatGPT, Gemini, Copilot, Cursor or any in-house agent at /api/v2/mcp and instantly access 290 warehouse operations — wave release, slotting, labor planning, robotics orchestration, inventory, shipping and analytics — through an open standard. No custom code, no proprietary SDK, no vendor lock-in on the AI layer. Your WMS. Your model. Your choice.
What Is an AI MCP Server?
An AI MCP server is a service that exposes the capabilities of a business system — like a Warehouse Management System (WMS) — to any large language model through the open Model Context Protocol (MCP). Instead of every AI vendor writing a custom integration for every backend, MCP defines one protocol that every modern LLM already understands. The model connects once and can read data, run operations and stream results, with full audit and security on the server side.
MCP was originally published by Anthropic in late 2024 and is now backed by OpenAI, Google, Microsoft, Cursor and the broader AI ecosystem. It is to AI what USB-C is to hardware: one connector, one wire format, every device. The protocol runs over JSON-RPC 2.0, transported as Server-Sent Events for HTTP or stdio for local processes — the same standard that lets Claude Desktop read your filesystem, Cursor query your codebase, and reference agents talk to GitHub. JASCI now extends that surface to your live warehouse.
The three primitives an MCP server exposes
Tools
Callable functions with JSON Schema input contracts. JASCI exposes every operational capability — release_wave, optimize_slotting, get_inventory_summary, assign_picker, dispatch_amr — as an MCP tool the LLM can invoke directly.
Resources
Read-only addressable data the model can pull on demand. JASCI publishes its knowledge store, setup guides, FAQs and domain concepts as MCP resources so any LLM can ground answers in real platform documentation instead of hallucinating.
Prompts
Server-managed prompt templates. JASCI publishes battle-tested system prompts for wave coordination, labor planning, slotting analysis and 20+ other warehouse roles — your model walks in already an expert.
How Does an AI MCP Server Work?
Under the hood, the JASCI MCP server is a thin protocol adapter on top of the same AI Tool Registry that powers JASCI's internal copilots. Every operational capability in the platform — from release_wave to optimize_slotting — is already a structured tool with a JSON Schema input contract. MCP simply exposes them to the outside world over the open standard.
End-to-End Call Flow
Client Connects
Claude Desktop, Cursor or any MCP-compatible agent opens an HTTPS connection to /api/v2/mcp, sending a Bearer token. The JASCI auth filter resolves the token to a tenant, company, fulfillment center and role.
Tool Discovery
Client issues tools/list. JASCI returns all 290 tools as MCP descriptors with name, description, JSON Schema input contracts and required permissions. The LLM now knows everything the warehouse can do.
Natural-Language Request
User asks the LLM: "Release any orders headed to UPS that are within 30 minutes of cutoff and rebalance pickers in Zone B." The LLM plans the call sequence using JASCI tool descriptors.
Tool Invocation
Client issues tools/call. JASCI validates arguments against the JSON Schema, applies tenant scope, runs the tool, writes audit and usage logs, and returns the result. Streams over Server-Sent Events for long-running operations.
Full Audit Trail
Every call is logged with source=MCP, the API key id, the user, the tenant, the input arguments, and the output. Write operations also write a config-change audit row. Forensic-grade observability, day one.
A real example, end to end
A warehouse manager opens Claude Desktop and asks: “What's the cycle-count backlog in Building 2, and can you assign the four oldest discrepancies to the night shift?”
That is the entire architecture. A single open endpoint. One JSON-RPC dialect. Any LLM client. Every operation an enterprise WMS can perform, available to the AI agent of your choice — with the same tenant security, role checks and audit trail that protect the production UI.
What Are the Benefits of an AI MCP Server?
An MCP server collapses the historic integration tax between AI vendors and enterprise systems. Where customers used to need a separate point-to-point integration for every model — and a separate rip-and-replace project every time procurement chose a new AI provider — MCP turns AI interoperability into a one-time architectural decision that pays compounding dividends for years.
One protocol, every LLM
Connect Claude, ChatGPT, Gemini, Copilot, Cursor, or any open-source LLM to the same server with no per-model integration work. New models become useful the day they support MCP.
Zero AI vendor lock-in
Swap models freely as cost, capability or compliance evolves. Your prompts, workflows and audit trail are tied to the server — not to a specific model provider that can change pricing overnight.
Faster time-to-value
No custom SDK, no proprietary client library, no months-long integration project. A single JSON config block in any MCP-compatible client unlocks the full capability surface.
Centralized governance
Tenant scope, RBAC, schema validation, rate limiting and audit live on the server side. The model never sees what it is not entitled to see — and every action it takes is logged.
Multi-LLM routing
Use the right model for each job — long-context reasoning, vision, voice, cost-optimized batch — all hitting the same tool catalog. Run several LLMs in parallel against the same backend.
Bring-your-own-model
Connect self-hosted Llama, Mistral or fine-tuned private models through MCP-compatible runtimes. Sensitive data and regulated workloads never have to leave your perimeter.
Compounding ecosystem effects
Every new MCP-compatible client that ships in the ecosystem instantly works with your server. The value of the MCP investment grows automatically as adoption widens.
Natural-language operations
Operators stop clicking through dozens of screens. They describe the outcome and the LLM orchestrates the right tool calls — speeding routine work and lowering training burden.
Future-proof architecture
Building on an open standard means new agent frameworks, IDE assistants, voice copilots and orchestration platforms plug in automatically as the ecosystem evolves.
The bottom line: an AI MCP server turns “which AI vendor should we bet on?” from a high-stakes, multi-year procurement decision into a portfolio choice you can revisit every quarter. The system of record stays where it belongs. The model becomes a swappable component — exactly what an enterprise AI strategy should look like in 2026.
The Announcement: Open AI Interoperability Comes to the Warehouse
In August 2026, JASCI will go live with the industry's first standards-compliant Model Context Protocol (MCP) server for warehouse management. Every JASCI tenant will be able to provision API keys and connect their preferred LLM — Anthropic Claude, OpenAI ChatGPT, Google Gemini, GitHub Copilot, Cursor, or any in-house agent — directly to the live warehouse, with full tenant security and audit trail.
This is the moment the warehouse industry stops being a black box for AI. For two decades, WMS vendors have promised “AI-powered” features inside locked-in agent platforms — you had to use their model, their UI, their runtime, and pay their margin on every token. JASCI takes the opposite path. The platform exposes the same 290 operational tools our internal agents use, over an open protocol that already runs in Claude Desktop, Cursor, and every major LLM SDK. The model is your choice. The data and the system of record stay where they belong — in JASCI.
Open standard
Model Context Protocol — adopted by Anthropic, OpenAI, Google, Microsoft and Cursor.
290 tools
Every JASCI operation surfaced through one secure endpoint, no custom integration.
Enterprise-grade
Tenant-scoped keys, RBAC, schema validation, per-call audit and rate limiting.
Why Vendor Lock-in Is the Hidden Tax on Enterprise AI
Most WMS vendors talk about AI but ship it as a closed black box. Typical WMS systems are a walled garden where the model, orchestration layer, UI, and billing are all tied together. Once committed, customers cannot swap models, cannot bring private data into a private LLM, and cannot use the AI vendor of their corporate AI committee. This is the same lock-in problem that ERP vendors created in the 2000s, replayed at the AI layer.
Closed Agent Platforms
- One model, chosen by the vendor — usually their margin model
- Can't run procurement-approved LLMs (Claude, ChatGPT, Gemini Enterprise)
- No way to plug into your in-house copilot or internal agent
- Tokens are billed by the WMS vendor at a 3-5x markup
- Compliance team has to re-approve every model change
- Lift-and-shift to a new vendor means rebuilding every AI workflow
JASCI Open MCP
- Any LLM — Claude, ChatGPT, Gemini, Copilot, in-house, open-source
- Procurement just adds JASCI as an MCP source; no new AI vendor
- Plug straight into existing internal copilots and agent stacks
- You pay your AI provider directly at retail; JASCI takes zero margin
- Swap models freely as policy or capability changes
- Multi-LLM strategies — Claude for reasoning, ChatGPT for code, Gemini for vision
Five LLMs, One Platform: True AI Interoperability in Action
Because JASCI speaks the open Model Context Protocol, every major LLM ecosystem can talk to the warehouse natively — no shim, no adapter, no custom integration. Customers can run multiple LLMs in parallel, routing different workloads to the model best suited for each task.
Anthropic Claude
Long-context reasoning, wave coordination, exception triage. Connect via Claude Desktop or the Anthropic SDK.
OpenAI ChatGPT
Natural-language operator interfaces, voice copilots, customer-facing chat. Connect via the OpenAI Responses API.
Google Gemini
Multimodal analytics, image-based receiving validation, dashboard generation. Connect via Gemini with MCP support.
GitHub Copilot
IT and integration teams calling JASCI from inside their dev workflow to debug data, generate reports and prototype agents.
Cursor
Engineering teams building custom workflows, integrations and agents against JASCI without leaving their editor.
Your Private LLM
Self-hosted Llama, Mistral or fine-tuned model. Any MCP-compatible runtime can connect — your model, your infrastructure.
Claude → ChatGPT → Gemini → Copilot → Your LLM — same MCP endpoint, same 290 tools, zero code change.
What Can Your LLM Actually Do? The 290 Tool Catalog
The JASCI MCP server exposes every operational capability in the platform. From day one, your LLM can read live data, run optimizations, dispatch robots, assign labor, release waves, and generate operator-grade reports — all under the API key's tenant scope and role permissions.
| Domain | Tools | Example LLM Capability |
|---|---|---|
| Wave Management | 38 | "Release any FedEx orders within 45 min of cutoff." |
| Slotting Optimization | 24 | "Suggest a re-slot for the top 50 movers in Zone C." |
| Labor & Workforce | 31 | "Reassign two pickers from Zone A to Zone B for the next hour." |
| Robotics Orchestration | 22 | "Dispatch an AMR to dock 7 for the inbound pallet." |
| Inventory Control | 47 | "What's on-hand for SKU 91823 by lot and expiration?" |
| Order Fulfillment | 41 | "Show me any orders that missed their SLA today and why." |
| Receiving & Putaway | 29 | "Pre-build a putaway plan for the 11 AM Costco trailer." |
| Cycle Counting | 18 | "Schedule a blind count for the 12 highest-variance locations." |
| Shipping & Carrier | 26 | "What's the current carrier-cutoff exposure for the next 90 min?" |
| Analytics & Reports | 14 | "Build a weekly throughput report for Building 2." |
As JASCI ships new capabilities, they automatically appear in the MCP catalog the moment they are registered with the AI Tool Registry. There is no client-side update, no SDK re-publication, and no version negotiation. The platform evolves; your LLM stays current.
What Makes JASCI Truly AI-Native — Not Just AI-Decorated
“AI-native” is one of the most abused terms in supply-chain marketing. The MCP server is not a feature bolted onto a legacy WMS — it is the natural surface of a platform architected from the ground up for autonomous agents. Five design choices make this possible.
Tool-First Architecture
Every JASCI operation — internal or external — flows through the same AiTool interface. The same code that powers the in-product copilot powers the MCP server. There is no "AI surface" and "real surface" — they are one and the same.
Structured Tool Registry
Every tool ships with a JSON Schema input contract, a description, a required-permission declaration and a context tag. The registry is the canonical contract for both human-written code and LLM-driven invocations.
Operational Knowledge Store
JASCI publishes 200+ structured knowledge entries — setup recipes, FAQs, domain concepts, troubleshooting guides — as MCP resources. Your LLM grounds answers in real platform documentation instead of hallucinating.
Audit-First Tool Execution
Every tool call writes a usage log row and, for writes, a config-change audit row. Tenants get a forensic-grade trail of every action an LLM ever took on their data, queryable from the admin console.
Streaming-Native Transport
JASCI already runs Server-Sent Events for analytics toasts and waveless release. The MCP transport reuses the same battle-tested pattern. Long-running optimizations stream incremental results back to the model — no timeouts, no polling hacks.
Enterprise-Grade Security From Day One
Opening the WMS to outside LLMs is only credible if the security model is uncompromising. The JASCI MCP server is built on the same identity, tenant and audit primitives that already protect the production UI and API — extended with explicit safeguards for autonomous agent traffic.
| Concern | Mechanism |
|---|---|
Authentication | Bearer API key, stored hashed (argon2) at rest. Plaintext shown once at creation. Per-key expiration and revocation. |
Tenant scoping | Every API key is bound to a tenant, company and fulfillment center. Tools execute in that scope — cross-tenant access is structurally impossible. |
Authorization | Role-based: Viewer / Team Member / Manager / Admin. Destructive tools (delete_*, void_*, bulk_*) require admin scope and explicit preflight tokens. |
Input validation | JSON Schema validation against the tool contract before execution. Invalid input returns JSON-RPC -32602 with no side effects. |
Rate limiting | Per-key and per-tenant rate buckets. Configurable token-bucket policy with burst allowances for legitimate agent workflows. |
Audit trail | Every call writes an AI usage log row with source=MCP. Every write tool also writes a config-change audit row with the API key id, user and arguments. |
Transport security | HTTPS only, TLS 1.3 minimum. SSE keep-alive 30s, idle timeout 5 min. CORS disabled — server-to-server only. |
Prompt-injection defense | Destructive tools require admin keys. Preflight tokens for irreversible operations. Tenant-scoped data access. Audit trail enables forensic reversal. |
What Will Customers Build on Top of the JASCI MCP Server?
The following sketches describe five repeatable patterns enterprises are evaluating. Each starts by pointing an existing LLM client at the JASCI MCP endpoint.
The Bring-Your-Own-Copilot Pattern
A Fortune 100 retailer has a corporate-approved Claude Enterprise tenant with strict data residency rules. Their warehouse team adds JASCI as an MCP source. Operators talk to the existing Claude UI — same authentication, same compliance posture — but the model can now query inventory, release waves and rebalance labor in real time. Zero net-new AI vendors, zero new procurement cycles.
The Multi-LLM Routing Pattern
A 3PL routes different workloads to different models: Claude for long-context wave planning, ChatGPT for voice-based operator chat, Gemini for vision-based receiving validation, and a self-hosted Llama for after-hours batch jobs. All four models share the same JASCI tool catalog through MCP. Cost, capability and compliance optimized per workload.
The Embedded-Agent Pattern
A high-growth D2C brand builds its operations dashboard in-house. The dashboard embeds a customer-built copilot using the OpenAI Agents SDK and registers the JASCI MCP server as a tool source. Operations leaders ask questions in plain English and the copilot executes against real warehouse data with full audit trail — without the brand building a single warehouse-specific tool from scratch.
The Robotics Concierge Pattern
A robotics-heavy fulfillment center pairs JASCI's orchestration tools with Cursor on the engineering floor. When a fleet anomaly occurs, an engineer asks Cursor to pull dispatch history, recent firmware events and zone load — Cursor calls the JASCI MCP tools, correlates the data and suggests a remediation. Mean-time-to-resolution drops from hours to minutes.
The Private-LLM-On-Sensitive-Data Pattern
A defense logistics customer runs a fully on-premises fine-tuned model. The model connects to JASCI through MCP over a private network link. No data ever leaves the customer's perimeter, no calls are made to a public LLM provider, and every tool invocation is logged in JASCI's audit trail for compliance review.
Open MCP vs. Typical Bundled WMS AI
Every WMS vendor will eventually claim “agentic AI.” The question is whether their architecture lets you own the AI strategy or forces you to inherit theirs. This table contrasts JASCI with a generic pattern we see in many legacy WMS plus packaged AI — not any single named product.
| Capability | JASCI | Typical bundled WMS AI / agent model |
|---|---|---|
| Open MCP server | ✓ Aug 2026 | Rarely: an open MCP surface over the full WMS operational tool catalog |
| Choose your LLM | Any | Often constrained to vendor-bundled or approved models |
| Bring your own private LLM | ✓ | Usually not a first-class, supported pattern |
| Multi-LLM routing | ✓ | Uncommon as a promoted architecture |
| Pay AI provider directly | ✓ | Often billed or bundled through the WMS vendor |
| Operator agent connectivity | Any MCP client | Proprietary connectors, studios, or sandboxes more common than MCP-first |
| Full operational tool catalog | 290 tools | Agent scope often a curated subset of WMS capabilities |
| Per-call audit + tenant scope | ✓ | Varies — common in enterprise-grade WMS |
Illustrative architectural comparison — not exhaustive. Individual vendors and editions differ; evaluate each roadmap and contract against your AI and compliance requirements.
How to Connect Your LLM to JASCI in Under 10 Minutes
When the MCP server goes live in August 2026, getting started will take less time than booking a demo. Four steps, one config block.
Four-Step Quickstart
Provision an MCP API key
In the JASCI admin console, open Settings → API Keys → MCP. Click "Create key", choose a role and scope (tenant, company, fulfillment center), and copy the plaintext key. The key is shown once.
Register the JASCI MCP server in your LLM client
In Claude Desktop, Cursor, or your custom agent, register the JASCI endpoint and the API key. The exact format varies by client but is always a single JSON config block.
Verify tool discovery
Restart the client. It will automatically call tools/list and discover all 290 JASCI tools. Try a safe read-only request like "What was today's pick throughput in Building 1?"
Roll out to operators
Issue scoped keys per role. Start with read-only viewer keys, expand to manager keys for routine operations, and reserve admin keys for senior staff. Monitor the audit log in real time.
Reserve Your Spot in the August 2026 Launch
Get in touch about the August 2026 launch, choose your LLM, and see what AI-native warehouse management actually feels like — with the WMS that lets you own the model.
Frequently Asked Questions
Key Takeaways
Craig Wilensky is the founder and CEO of JASCI Software, the AI-native warehouse management platform. With more than 20 years of experience in supply chain technology, Craig has led the development of JASCI's AI Tool Registry, continuous order release, AI-powered slotting and robotics orchestration — and now the industry's first open MCP server. He works directly with enterprise 3PLs, retailers and D2C brands to design WMS architectures that give customers full ownership of their AI strategy.
