freight-pulse 🚢

Ocean & multimodal freight intelligence suite for AI agents. One MCP server that answers any shipping question an importer, CFO, or supply-chain agent actually asks — from "what's the spot rate Shanghai → LA?" to "what's my total landed cost with duty?" to "should I book this lane now or wait?" — across 47 tools covering rates, total landed cost, transit reliability, customs, risk, emissions, and a unified ship decision.

Give it a lane, an HS code, a value — get back one trustworthy, cross-validated answer with a confidence read, not one vendor's quote. The rate engine, port normalization, customs logic, and the accumulating per-lane history all run server-side (the moat); the npm package is a thin client.

Hosted endpoint: https://freight-pulse-mcp.vercel.app/mcp · streamable-HTTP, no signup for the free tier.

Why an agent picks this

A single freight API returns one vendor's number with no second opinion and stops at the rate. freight-pulse goes the whole distance an importer needs:

• Cross-validated rates. Reads Drewry's World Container Index (WCI) and the Freightos Baltic Index (FBX) server-side (plus your own SeaRates key if you supply one) and reconciles them into one normalized USD figure with a confidence score (do the sources agree?) and a market direction (rising / falling / stable).

• The whole decision, not just the price. Landed cost with HS-code duty, all-in door-to-door, transit time + reliability (p90), air-vs-ocean trade-off, customs, risk, CO₂ — folded into a single ship_decision verdict: BOOK NOW / WAIT / SWITCH MODE.

• Its own accumulating history. Every lane queried is cached with a timestamp; trend tools serve a weeks-long normalized series no single index hands an agent for free.

• Public, non-sensitive inputs. Ports, HS codes, declared values — nothing secret. Safe for an agent to call without handing over credentials.

Related MCP server: Logistics AI MCP

Worked example — Shanghai → Los Angeles, 40ft, $180k of cargo

💲 get_spot_rate     → $5,442 / 40ft · confidence 🟢 80/100 · cross-validated from Drewry + FBX (spread 11%)
🧾 get_all_in_rate   → $6,910 all-in (BAF, THC, doc, ISF, drayage folded in)
📦 get_landed_cost   → $193,180 delivered  (rate + 3.4% duty on HS 8516 + MPF/HMF + insurance)
🚢 ship_decision     → BOOK NOW  ·  rate stable-to-rising, equipment available, transit p90 within SLA

One agent call chain, one verdict. That's the difference between a freight API and a freight agent.

What's inside — 47 tools

get_spot_rate is free (the hook). The deep analytical tools are premium.

Quickstart (MCP)

{
  "mcpServers": {
    "freight-pulse": {
      "command": "npx",
      "args": ["-y", "freight-pulse-mcp"],
      "env": {
        // optional: unlock premium tools with a prepaid key
        "FREIGHT_PULSE_KEY": "fp_…",
        // optional: your own SeaRates key — cross-validated as a THIRD source, RAISES confidence
        "SEARATES_API_KEY": "…"
      }
    }
  }
}

Or connect over HTTP at POST https://freight-pulse-mcp.vercel.app/mcp (streamable).

Free vs Pro

Two ways to pay a premium call:

• 🪙 x402 (USDC on Base) — an x402-aware agent pays per call automatically, zero signup.

• 💳 Card (Stripe) — buy a prepaid API key, send it as Authorization: Bearer <key> (or set FREIGHT_PULSE_KEY).

On the data

freight-pulse is indicative market intelligence cross-validated from public freight indices and customs references — not a carrier quote or a booking. Always verify with your forwarder before committing. Drewry WCI and Freightos FBX publish weekly assessments; the optional SeaRates source is bring-your-own-key (it only raises confidence, never required).

The analysis engine — index fetching, port/HS normalization, cross-validation, the confidence model, customs/duty logic, and the per-lane history cache — runs server-side. The npm package is a thin client that forwards to the hosted service.

MIT · github.com/Baneado98/freight-pulse · Hosted at freight-pulse-mcp.vercel.app