Flash Warehouse MCP Server Goes Live: The First Chinese WMS to Support Model Context Protocol
Flash Warehouse launches its MCP Server, becoming the first Chinese WMS to support Model Context Protocol. Through the fwh CLI tool, AI agents like Claude Code, Cursor, and Windsurf can directly query and operate your warehouse data with 110 tools covering the full inventory lifecycle.
A New Milestone
Today, Flash Warehouse WMS officially launches its MCP Server capability. To our knowledge, this makes Flash Warehouse the first Chinese warehouse management system to fully support the Model Context Protocol.
What does this mean in practice? Starting today, AI agents can directly read and operate on your warehouse data -- check inventory, create purchase orders, audit bills, generate business reports -- all through natural language, without opening a single web page.
What Is MCP
Model Context Protocol is an open standard proposed by Anthropic that defines how AI models communicate with external tools. In simple terms, MCP lets an AI agent "use software" the way a human does.
Flash Warehouse's MCP Server is delivered through fwh, a cross-platform CLI tool written in Go. It compiles to a single static binary with zero runtime dependencies, supporting macOS, Linux, and Windows. It serves dual duty: a traditional command-line tool with 23 top-level commands, and a full MCP Server exposing 110 tools.
110 Tools Covering the Full Inventory Lifecycle
The MCP tools are organized by business domain, achieving full feature parity with the PC web frontend:
- Goods management: list, get, barcode search, create, modify, delete
- Warehouse management: list, create, modify, delete
- Stock alerts: overview, list, config, refresh
- Bill operations: full CRUD across all 16 bill types, plus audit, reject, and 7 bill conversion flows (inquiry to order, order to receipt, and more)
- Business intelligence: 8 BI tools including total inventory value, I/O trends, daily sales revenue, and receivables/payables
- People and permissions: partners, staff, roles, and the complete RBAC permission matrix
- POS: barcode-based retail sales via
pos_sell - Stocktake: create, execute, and finish inventory check tasks
Of these, 44 are read-only tools (queries), always available by default. The remaining 66 are write tools (create, modify, delete, audit), which require explicit opt-in.
Three Major AI Agents Supported
Flash Warehouse MCP Server has been verified with:
Claude Code -- configure in ~/.claude/mcp.json:
{
"mcpServers": {
"flash-warehouse": {
"command": "/path/to/fwh",
"args": ["mcp"]
}
}
}
Cursor -- same JSON format in ~/.cursor/mcp.json.
Windsurf -- same format in ~/.codeium/windsurf/mcp_config.json.
To enable write operations, change args to ["mcp", "--enable-writes"].
The Safety Model: Four Layers of Protection
Connecting an AI agent to a production system demands rigorous safety guarantees. Flash Warehouse MCP Server is built around four hard invariants:
Login-first. The MCP Server will not start without an active session. You must run fwh login in your terminal before launching the server. Agents cannot log in via tool calls -- this is intentional, preventing plaintext passwords from appearing in conversation history.
Tenant-locked. The binding_user_id that scopes all data is set once at login time and stored in an unexported Go struct field with a getter and no setter. No CLI flag, environment variable, or MCP tool argument can override it. A source-level audit test (TestTenantLock_UserIDChokepoints) enforces this invariant on every build.
Writes are opt-in. Without --enable-writes, write tools are simply not registered in the MCP Server's tool registry. An agent cannot call a tool that does not exist. Every write tool's description begins with WARNING: so the model's pre-flight reasoning flags the operation's destructiveness before execution.
Session revocation. If fwh logout is executed while the MCP Server is running, the server detects the missing session file on the next tool call and returns a session_revoked error immediately.
Get Started in Three Steps
Step 1: Build fwh
cd flash_warehouse_cli
go build -o ./bin/fwh ./cmd/fwh
ln -sf $(pwd)/bin/fwh ~/.local/bin/fwh
Step 2: Log in
fwh login --user your_phone_number
# Enter password when prompted
fwh whoami # Verify your identity
Step 3: Configure your agent
For Claude Code, edit ~/.claude/mcp.json:
{
"mcpServers": {
"flash-warehouse": {
"command": "/absolute/path/to/fwh",
"args": ["mcp", "--enable-writes"]
}
}
}
Once configured, you can interact with your warehouse through natural language in your AI agent. For example:
- "What is the current stock level of Coca-Cola in the main warehouse?"
- "Create a purchase receipt for 500 cases of mineral water from Supplier Wang into Warehouse 2"
- "Show me the input/output trends for this month"
- "Approve bill PO-20260601-001"
What Comes Next
The current fwh is at v0.2 with full PC feature parity. On our roadmap:
- GoReleaser-based release pipeline with
brew installandscoop installsupport - SSE streaming for AI tool responses
- Verification with additional agent platforms
Flash Warehouse remains free and open. If you manage a small or medium-sized warehouse and use AI coding tools, you can now experience a fundamentally new way to manage inventory through natural language.
Visit flashwarehouse.cn to learn more, or build fwh from source and start using it today.