Deal Report · $4.99Sign in

Give your agent AutoArb intelligence

Search the live directory, open a listing’s economics, and carry every unknown into the recommendation. Use the MCP bridge inside Codex or Claude Code, or call the same API from the terminal.

01 · Account

Create one revocable key.

  1. Sign in and activate Pro.
  2. Open Settings, choose Agent access, and create an API key.
  3. Save the secret when it appears. AutoArb shows it once and stores only its hash.

Keep the key in your environment. Do not paste it into a prompt, commit it, or place it in a shared MCP file.

02 · Terminal

Check access before involving an agent.

The CLI is the fastest way to verify the URL, key, subscription, and current directory.

export AUTOARB_API_KEY="aak_live_your_key"
export AUTOARB_API_URL="https://autoarb.vercel.app"

CLI="npx -y --package=github:EnzoVezzaro/autoarb autoarb"

$CLI doctor
$CLI stats
$CLI search "2021 Ford Bronco" --max-price 32000 --limit 5
$CLI listing <listing-id> --json
03 · MCP

Connect the local bridge.

The bridge uses stdio and forwards requests to AutoArb’s hosted API. It never places bids or purchases vehicles. Use the package command for a zero-path setup, or the local clone if you already have the repository.

Codex

# ~/.codex/config.toml — zero-path setup
[mcp_servers.autoarb]
command = "npx"
args = ["-y", "--package=github:EnzoVezzaro/autoarb", "autoarb-mcp"]

[mcp_servers.autoarb.env]
AUTOARB_API_URL = "https://autoarb.vercel.app"
AUTOARB_API_KEY = "aak_live_your_key"

Already cloned? Use node <path-to-autoarb>/mcp/server.mjs instead of the npx command.

Claude Code

// .mcp.json
{
  "mcpServers": {
    "autoarb": {
      "command": "npx",
      "args": ["-y", "--package=github:EnzoVezzaro/autoarb", "autoarb-mcp"],
      "env": {
        "AUTOARB_API_URL": "https://autoarb.vercel.app",
        "AUTOARB_API_KEY": "aak_live_your_key"
      }
    }
  }
}

Local clone alternative: node <path-to-autoarb>/mcp/server.mjs.

04 · Skill

Teach the agent how to use the evidence.

The portable skill defines the search-to-verification workflow, quota behavior, and the fields every recommendation must disclose.

# Codex
cp -R skills/autoarb-agent ~/.codex/skills/autoarb-agent

# Claude Code
cp -R skills/autoarb-agent ~/.claude/skills/autoarb-agent
05 · Verify

Ask for a result you can audit.

Find five buyable 2021–2023 Ford Broncos below $32,000.
Open every candidate you might recommend. For each one, report the
maximum acquisition price, expected profit, source URL, snapshot time,
and every unknown that could change the decision. Do not recommend a
bid when the evidence is too thin.
search_listings

Search by vehicle, source, tier, price, year, mileage, or result limit.

get_listing

Open one candidate’s ceiling, economics, source evidence, and unknowns.

directory_stats

Check directory size, source coverage, and snapshot time before broad claims.

Decision boundary

The agent advises. You decide.

AutoArb has no automatic bidding or purchasing pathway. It gives your agent the arithmetic and provenance needed for a decision you can check.

Agent setup — AutoArb