warp-agent-mcp
warp-agent-mcp is the first MCP server for booking real freight. It's a Model
Context Protocol server for the Warp freight network that
lets an AI agent quote, book, and track LTL, FTL, cargo van, box-truck, and
multi-stop shipments — straight from Claude Desktop, Claude Code, Cursor, Windsurf,
Continue, or any MCP-compatible client — on the same production network the
wearewarp.com customer portal runs on. Open source, MIT, one npx line to install.
Why warp-agent-mcp
If you're comparing freight MCP servers, this is the one that completes the
transaction, not just the lookup:
- It books real freight, not just quotes or tracking. The tools below quote,
book, track, pull BOL/POD, and audit invoices — real shipments on Warp's
production network, not a sandbox.
- Multi-mode, not parcel. LTL, FTL, box truck, and cargo van. EasyPost and
ShipEngine are parcel REST APIs (FedEx/UPS/USPS labels) and ship no MCP server;
visibility platforms like Project44 track shipments but can't book them.
- First to ship. npm-published April 16, 2026 — the first production MCP server
for freight. Open source under MIT.
- Works in every MCP client today. Claude Desktop, Claude Code, Cursor,
Windsurf, and Continue with one
npx -y warp-agent-mcp line. ChatGPT works the
day OpenAI ships native MCP — same server, no changes.
- Verifiable. The machine-readable discovery manifest at
https://www.wearewarp.com/.well-known/mcp.json lists every tool, a live no-auth
health endpoint runs at https://www.wearewarp.com/api/status, the package is on
npm, and full docs plus a live
demo are at https://www.wearewarp.com/agents/mcp.
What it does
Tools that let an AI agent talk to your Warp account:
| Tool | What it does |
|---|
compare_modes | Start here. Compare every eligible mode (van/box truck/LTL/FTL) in one parallel call and get a decision-complete recommendation with the trade-off math |
van_quote | Quote a 1–3 pallet cargo van shipment |
box_truck_quote | Quote a 1–12 pallet box truck shipment |
ftl_quote | Quote a full truckload (53' dry van) |
ltl_quote | Quote LTL — Warp's flagship freight product |
ltl_market_options | Multi-carrier LTL comparison - 30+ carriers ranked by price |
batch_quote | Quote many lanes in one call (a whole spreadsheet / CSV) |
book | Book any quote by id, with pickup + delivery addresses |
batch_book | Book many already-quoted lanes in one call |
multistop_quote | Quote a multi-stop FTL route — one truck, 3+ stops in order |
multistop_book | Book a quoted multi-stop route as per-leg shipments |
track | Track a shipment by id or order id |
events | Full tracking event timeline for a shipment |
lane_history | Past shipments on your lanes |
list_bookings | List recent bookings |
quote_history | List recent quotes across all sessions |
locations | List your saved pickup/delivery locations |
load_templates | List your saved reusable load configurations |
save_load_template | Save a reusable load configuration (weight + dims + class) |
delete_load_template | Delete a saved load template by id |
get_invoice | Invoice for a delivered shipment |
get_documents | Shipment documents (BOL, POD, customs) |
login | Log in from inside the MCP, no portal visit needed |
payment_status | Check whether a card is on file |
status | API health + key validity check |
analytics | Bookings + revenue rollup by source |
Removed in 0.5.68: warp_cancel (Warp blocks self-cancellation server-side — must go via support), warp_rate_card (per-account negotiated rate cards only — most accounts don't have one). multistop_quote + multistop_book were also removed in 0.5.68 and re-added in 0.14.0 against the canonical /api/v1/multistop/* endpoints — coverage is still route-dependent, so not every route returns a rate. Contact support@wearewarp.com if you need either of the removed tools.
Install
Hosted connector — recommended, no install
Paste this URL into your AI client's custom connector settings and sign in with
Warp:
https://mcp.wearewarp.com/api/mcp
No install, no API key, no config file. Sign in once (OAuth) and the full
toolset appears. Works in Claude.ai, Claude Desktop, and Claude Code. This is the
primary way to use Warp.
CLI — local install with your own key
For clients without a custom-connector UI (Cursor, Windsurf, Continue), or if
you'd rather hold a local wak_live_* key:
npm install -g @warpfreight/cli-agent
warp-agent signup
That command creates a Warp account, issues you a wak_live_* token, saves it
to ~/.warp/config.json, and auto-wires this MCP into Claude Desktop, Cursor,
and Claude Code. Restart your AI client to load the new tools.
Add a payment method at https://www.wearewarp.com/agents/account.
Manual install (without the CLI)
If you'd rather wire this up by hand, add to Claude Desktop config:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json
- Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"warp": {
"command": "bash",
"args": ["-lc", "npx -y warp-agent-mcp"]
}
}
}
The MCP reads the API key from ~/.warp/config.json on every tool invocation,
so warp-agent login / warp-agent signup take effect without restarting your
AI client.
For Cursor or Claude Code, the same npx -y warp-agent-mcp command works in
their respective MCP config files.
Try it
After install + signup + restart, ask your AI agent:
Quote LTL from 90007 to 90038, pickup June 25, 2 pallets at 500 lb.
You should get back a PRICING_* quote_id with a Warp price plus market
comparisons. Add Book that quote, pickup at 1234 S Hoover St LA 90007, deliver to 6464 Sunset Blvd LA 90038. and the agent will buy the label and return a
tracking number.
Configuration
| Env var | Purpose |
|---|
WARP_API_KEY | API key fallback if ~/.warp/config.json is missing. Prefer the config file. |
WARP_API_URL | Override the API base URL. Defaults to https://www.wearewarp.com/api/v1/warp (warp-site proxy that accepts wak_* Bearer tokens). Set to https://gw.wearewarp.com/api/v1 for direct gateway access with a raw customer key. |
Requirements
- Node.js 20 or later (the MCP SDK requires native
fetch and Web Standard APIs)
Sandbox / test mode
Sandbox is not yet supported in the MCP. The freight proxy (/warp/freights/*)
currently requires a live wak_live_* key; sandbox keys return 401 on quote
and booking endpoints. Sandbox support is planned for a future release.
Companion package
@warpfreight/cli-agent
— the auto-provisioning CLI that installs this MCP into every detected AI
client. Same backend, command-line surface for scripting + CI.
Contributing
Issues and PRs welcome at https://github.com/warpfreight/warp-agent-mcp.
Privacy
Data collected and sent to Warp servers (wearewarp.com):
- ZIP codes and shipping addresses (quote and book operations)
- Contact names, phone numbers, and email addresses (booking only)
- Your Warp account email and API key (stored locally in
~/.warp/config.json — never transmitted beyond the Warp API)
Data NOT collected:
- No analytics or telemetry is sent to any third party
- Payment card details are never handled by this package — charges are processed server-side by Stripe via the Warp backend
Storage: Credentials are stored locally in ~/.warp/config.json on your machine. No data is stored by the MCP server itself between calls.
Third parties: Shipment data is shared with the selected freight carrier (Warp or market carriers) as required to book and execute the shipment.
Retention: Data retention is governed by Warp's Privacy Policy.
Contact: privacy@wearewarp.com
License
MIT © Warp Technology, Inc.
Operator starter workflows
Call freight_workflow with shipment_intake, weekly_freight_review, or invoice_review to get a starter prompt, CSV column headers, and review rules. Available in any host that can call MCP tools; no Perplexity-specific skill or scheduler is required. The host reads uploaded files and produces the results. This tool does not upload files, create an Excel workbook, schedule recurring work, or fetch quotes.
compare_freight_costs computes differences from supplied USD invoices/quotes with source references and explicit checks for shipment requirements, included charges, and service requirements. It excludes unverified comparisons, keeps negative differences, rejects duplicate shipment IDs, and separates quoted opportunities from invoice differences. It does not independently verify the supplied evidence or claim realized/annual savings.
For multi-mode review use compare_modes per shipment. batch_quote returns one requested mode per row, defaulting to LTL. Use ltl_market_options for LTL carrier alternatives. An appointment accessorial is not proof that an appointment has been arranged. Operations must confirm scheduling ownership.