Skip to content
MCP Server Directory logo
finance-ecommerce

Packrift Mcp

Public MCP (Model Context Protocol) server for the Packrift packaging-supplies catalog. Live at https://mcp.packrift.com/mcp

Reviewed overview

MCP 服务器概览

What is Packrift MCP?

Packrift MCP is a production Model Context Protocol (MCP) server that exposes Packrift’s exact-spec packaging catalog and AI-commerce tooling. It lets agents, AI clients, and partner directories discover packaging SKUs, confirm live price/inventory/shipping, and perform cart handoffs for packaging procurement. The primary hosted endpoint is https://mcp.packrift.com/mcp.

How to use Packrift MCP

  • For most integrations, point your MCP-capable client to: { “mcpServers”: { “packrift”: { “type”: “http”, “url”: “https://mcp.packrift.com/mcp” } } }.
  • Endpoints: POST /mcp (JSON-RPC tool calls), GET /mcp (SSE), GET /start (developer/agent start page), and /.well-known/mcp/server-card.json.
  • Quick developer startup: call initialize, then tools/list and tools/call for specific tools. Example curl sequences are included in the repo README.
  • Self-hosting (optional): Docker image ghcr.io/packrift/packrift-mcp:latest or build locally. For live Shopify-backed calls you must set SHOPIFY_PACKRIFT_TOKEN; discovery works without it.
  • Tracked partner flows: use /r/start/{source}, /r/config/{source}, /r/install/{source}/{target} to capture install-intent and usage telemetry.

Key features

  • Exact-spec procurement tooling tailored to packaging: dimension-based matching, pack calculators, and no-exact-match recovery.
  • Live Shopify Admin-backed actions: get_pricing, check_inventory, get_shipping_estimate (via draftOrderCalculate), create_cart_url/cart handoff.
  • Rich tool surface (streamable HTTP-friendly) including: find_packaging_for_item, search_products, get_product, get_pricing, check_inventory, get_cart_handoff_candidates, prepare_purchase_handoff, compare_alternatives, pack_calculator, inventory_status, get_reorder_link, get_bulk_quote_link, explain_no_exact_match.
  • Hosted, no-auth primary endpoint for discovery and agent use; container image for local development or self-hosted installs.
  • Extensive discovery/manifest surfaces (.well-known mcp configs, OpenAPI, AI plugin manifests, LLMS lists) and partner connector listings (Glama, MCP Marketplace, registry entries).
  • Telemetry and adoption tooling: tracked start/install/config URLs, usage snapshots, smoke tests and GitHub Actions for live checks.

Use cases

  1. Agent-assisted packaging selection: give item L×W×H and weight to find smallest-fit box or mailer.
  2. Reorder flows: confirm SKU price/inventory and produce reorder cart links for purchasers.
  3. Procurement workflows: obtain bulk-quote links and procurement copy/specs for non-exact matches.
  4. AI/agent integration: add Packrift as an MCP server for automated cart handoffs and measured commerce attribution.
  5. Directory & partner handoffs: generate tracked install/start links for partner onboarding and telemetry.
  6. Local development and testing of MCP clients using the published Docker image and local Worker emulation.

FAQ

  • Q: Is the hosted endpoint the recommended integration path? A: Yes. The canonical production endpoint is https://mcp.packrift.com/mcp and is the primary integration surface for buyers, agents, and directories.

  • Q: Do I need a Shopify token to use Packrift MCP? A: No for discovery and most MCP introspection. SHOPIFY_PACKRIFT_TOKEN is optional for discovery but required when self-hosting if you need live Shopify-backed catalog, pricing, inventory, shipping, and cart tool calls.

  • Q: How do clients add Packrift to their MCP client list? A: Add an mcpServers entry with type “http” and url “https://mcp.packrift.com/mcp” (see README examples). Some targets (e.g., cline) have tracked install URLs returning native streamableHttp configs.

  • Q: What guarantees about exact matches and substitutions? A: Packrift is optimized for exact-spec procurement. Agents should not treat nearby sizes/materials/strengths as exact substitutes—if no exact AI-approved match exists, use reorder, no-match explanation, or bulk-quote recovery flows.

  • Q: Can I run this locally or in CI? A: Yes. A Docker image and local dev instructions are provided. Smoke tests and GitHub Actions are included for scheduled live checks.

  • Q: What transport/stack does it use? A: Cloudflare Workers (production), TypeScript (strict), Hono, Zod, and Streamable HTTP for live tool/agent interactions.

  • Q: How are tool errors surfaced? A: Tool exceptions are returned per MCP spec as tool-level error content objects ({ content:[…], isError: true }); protocol-level problems return JSON-RPC errors.

For full installation, tool reference, and examples see the repository: https://github.com/Packrift/packrift-mcp and the hosted start page: https://mcp.packrift.com/start