The first logistics/WMS MCP server for AI agents. Rate shopping, inventory, orders, fleet tracking, AI forecasting.
CerebroChain MCP server has 12 tools with mixed quality. All tools are registered with explicit Zod schemas and descriptions visible in src/index.ts. Naming is generally clear (verb_noun pattern: get_, list_, search_, check_, lookup_, track_, compare_). However, several critical gaps reduce the score: (1) Parameter descriptions are minimal, most lack actionable format/constraint guidance (e.g., 'Search term (name, SKU, barcode)' doesn't explain valid length, character restrictions, or failure modes). (2) Output schemas are completely undocumented, no response field definitions are visible for any tool, forcing LLMs to infer what fields to expect. (3) Error handling is generic, all tools return text responses with minimal recovery guidance ('Error: {error}' pattern observed in compare_shipping_rates, search_inventory). (4) Three tools (compare_shipping_rates, get_platform_status, get_warehouse_locations) have trivial parameter descriptions or none at all. (5) No tool explicitly documents idempotency, rate limits, or prerequisites beyond 'Requires API key.' Average tool description is 120 - 180 chars (within baseline 194 chars), but descriptions lack the LLM-optimized specificity of A-tier tools. Composition is clean, each tool has one responsibility, and tool names correctly use action verbs. All required parameters have type definitions via Zod. However, output structure is nowhere documented, breaking the pattern:tool-chain pattern.
Get current stock levels for a specific inventory item by ID. Returns quantity on hand, reserved, available. Requires API key.
Compare shipping rates across multiple carriers (UPS, FedEx, USPS, DHL) for a package. Returns cheapest and fastest options. Free — no API key needed.
Get real-time fleet KPIs — active vehicles, utilization, fuel efficiency, maintenance status. Requires API key.
Get order statistics and KPIs — total orders, fulfillment rate, average value, trends. Requires API key.
Get detailed status and tracking for a specific order. Requires API key.
Check CerebroChain platform health and service availability. Free — no API key needed.
No output schemas documented. Tools return raw JSON without field definitions, forcing LLMs to infer structure and breaking tool-chain composition. For example, search_inventory returns results but no field names are documented, preventing downstream tools from knowing what IDs or references to extract.
Parameter descriptions lack actionable constraints. Examples: 'Search term (name, SKU, barcode)' does not specify length limits, character restrictions, or whether partial matches are supported. 'Item ID' does not describe format (UUID, alphanumeric, numeric). LLMs cannot validate input against undocumented constraints and will produce invalid requests.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
List warehouse storage locations with hierarchy (zones, aisles, bins). Requires API key.
List and search orders with filters. Returns order ID, status, items, total value. Requires API key.
List fleet vehicles with status, location, and capacity. Requires API key.
Look up an inventory item by its SKU code. Returns full item details including stock, location, and pricing. Requires API key.
Search inventory items with filters. Returns SKU, name, quantity, location, category. Requires API key.
Get real-time tracking for a shipment route — current location, progress, ETA, stops completed. Requires API key.
Error responses are generic and non-actionable. All tools return 'Error: {error}' without recovery guidance. Example: 'Error: Invalid SKU' does not tell the LLM whether to retry, search first, or ask the user. Per pattern:recovery-guide, errors must guide the next step.
Empty-parameter tools lack descriptions. get_platform_status and get_warehouse_locations accept no parameters but have minimal descriptions (40 - 50 chars). Descriptions should explain: What does it return? When should the LLM call it? Is it a discovery tool that reveals available options?
No pagination guidance in descriptions. search_inventory and list_orders accept page/limit parameters but tool descriptions don't explain: Are results sorted? What is the default order? If the user calls multiple times, will results be consistent? This breaks agent planning.
Authentication error handling is inconsistent. Some tools check isAuthenticated and return a message; others may not. There's no centralized error class or consistent format. A tool that fails due to missing API key should suggest how to set CEREBROCHAIN_API_KEY.