# Errand (full)

> Errand is commerce infrastructure for AI agents. It lets AI assistants and custom agents buy, book, and schedule from businesses through one integration, speaking the open agentic commerce protocols: ACP, UCP, AP2, x402, and MCP. Businesses keep control through policies, signed mandates, and spend caps.

Built by Errand Labs, Inc. Contact: hello@errand.dev
Site: https://errand-hq.pages.dev

## Product overview

Errand connects to a business's existing stack (Shopify, Stripe, Square, a POS, a booking system, or a custom REST API) and publishes an "agent surface" from it. No replatforming is required. The agent surface consists of:

1. A structured, machine-readable catalog feed kept in sync with the source system
2. A hosted MCP server exposing strict tools: search_catalog, get_quote, check_slots, book, buy
3. Checkout endpoints speaking ACP (OpenAI + Stripe) and UCP (Google)
4. AP2-style mandate verification: every purchase carries a cryptographically signed intent and cart mandate proving user authorization
5. A generated llms.txt and schema.org markup for the merchant's own website

## The five modules

### Agentic checkout
A single checkout endpoint agents can complete end to end: cart, tax, fulfillment, confirmation. Quotes resolve tax, fees, and fulfillment before capture. Payment settles through the merchant's existing processor; no new merchant account. Orders write back idempotently into Shopify, POS systems, or custom APIs.

### Scheduling and bookings
Agents book appointments, tables, and services against real availability with two-way calendar sync. Merchants define buffers, lead times, party sizes, blackout dates, and cancellation rules. Reschedule and cancel flows work without a phone call.

### Mandates, identity, and spend controls
AP2-style signed intent and cart mandates verified per transaction. Per-agent-platform rate limits and spend caps. Support for verified agent identity via Visa Trusted Agent Protocol and Mastercard Agent Pay tokens as networks roll them out. Immutable audit log: who asked, what was authorized, what was charged.

### Machine-readable storefront
Structured product and service feeds, hosted MCP server, generated llms.txt, schema.org markup, and answer-engine visibility reporting (where agents see the business and where they do not).

### Observability and analytics
Revenue by agent surface (ChatGPT, Gemini, Claude, custom agents), refusal analytics showing which requests failed policy, order-level traces from first query to settled payment, warehouse export.

## Control model

Every agent capability is a policy the merchant configures: which products are purchasable, order size limits, booking windows, refund rules, rate limits per agent platform. Requests outside policy are refused and logged. Example policy:

checkout: max_order_value 500 USD, require_mandate true
booking: window_days 30, cancellation_hours 24
platforms: chatgpt rate_limit 60/min, unverified allow quote_only

## Pricing

- Starter: $0 while in beta. Catalog + llms.txt, hosted MCP server, up to 200 agentic orders/month, community support.
- Growth: $249/month. Adds ACP/UCP/AP2 checkout, mandates and spend controls, scheduling, unlimited orders at 1% per agentic transaction.
- Platform: custom annual pricing for marketplaces and platforms embedding agent commerce for their merchants. Multi-tenant onboarding, custom protocol adapters, SLAs.

Early access partners keep beta terms for 12 months after general availability.

## Key definitions (from the Errand blog)

Agentic commerce: when an AI assistant completes a purchase or booking on a real customer's behalf, choosing the item, presenting proof of authorization, and paying, without the customer visiting the website.

Agent-ready: a business is agent-ready when AI assistants can discover it, quote it, and transact with it programmatically.

Protocol layers (mid-2026): ACP (OpenAI + Stripe) and UCP (Google) cover checkout and catalog; AP2 (Google + 60 partners) covers authorization via signed mandates; x402 (Coinbase) covers HTTP-native stablecoin payments; MCP (Anthropic, open standard) covers how agents discover and call tools. No single protocol has won; Errand abstracts all of them behind one integration.

## Market context (cited on the site)

- 68% of U.S. Google searches end without a click (Search Engine Land, 2026)
- Shopify: orders from AI search up ~13x year over year in Q1 2026
- Adobe: AI-driven traffic to U.S. retail sites up 693% YoY, 2025 holidays
- Gartner: machine customers to influence $30T in purchases by 2030; organic search traffic to fall 25% by 2026
- McKinsey: $3 to 5 trillion global agentic-commerce revenue by 2030
- Rutgers/Wharton (Dec 2025): publishers blocking AI crawlers lost 23.1% of total traffic without reducing citations

## Pages

- Home: https://errand-hq.pages.dev/
- Platform: https://errand-hq.pages.dev/platform
- Developers: https://errand-hq.pages.dev/developers
- Pricing: https://errand-hq.pages.dev/pricing
- Blog: https://errand-hq.pages.dev/blog
- Company: https://errand-hq.pages.dev/company
- Contact: https://errand-hq.pages.dev/contact