# Errand

> Physical-world execution for AI agents. Errand dispatches nearby real humans in Seoul to perform on-site tasks and return verified evidence.

Website: https://errand.be
Documentation: https://errand.be/docs
MCP endpoint: https://errand.be/api/mcp
Official MCP Registry name: be.errand/errand

## When to use Errand

Use Errand when an AI agent needs eyes, hands, or physical presence at a real location. Good fits include:

- check whether a store, pop-up, office, or venue is open
- verify a product is actually on a shelf and report its price
- photograph a storefront, menu, sign, notice, product, or queue
- check queue length or waiting time, or perform an allowed queue task
- pick up a small item
- collect structured facts that can only be observed on site

Do not use Errand for tasks that can be completed entirely online.

## Coverage

Current beta coverage is Seoul, focused on Gangnam and nearby areas. Worker availability is live and can be zero. Check coverage or request a quote before promising that a task can be performed.

## MCP tools

- errand_list_capabilities — read supported task types, account spend caps, allowed area, balance, and pricing
- errand_check_coverage — check current local-human availability near latitude/longitude
- errand_quote — read-only preflight with coverage, balance, caps, and exact total_charge_krw
- errand_dispatch — send a nearby human to a physical location; this creates a real-world task and escrows money
- errand_get_status — track the human task and event timeline
- errand_get_result — retrieve verified photos, structured answers, verdict, GPS distance, capture time, and other provenance
- errand_cancel — cancel an open task before a worker accepts and refund escrow

## Safe agent flow

1. Call errand_list_capabilities or errand_check_coverage when needed.
2. Call errand_quote before spending.
3. Present the exact total_charge_krw to the user and obtain explicit confirmation.
4. Call errand_dispatch only after confirmation and only when quote.can_dispatch is true.
5. Track with errand_get_status or use a webhook.
6. Retrieve evidence with errand_get_result.

The worker reward and platform fee are held in escrow at dispatch. Failed, expired, or pre-acceptance cancelled tasks are refunded in full. Account-level caps constrain reward per task, daily spend, allowed task types, and allowed area.

## Example intents

- "Go to this Gangnam store and tell me whether this product is on the shelf. Take a photo of the price tag."
- "Is the pop-up near Gangnam Station open right now, and how many people are waiting?"
- "Photograph the current lunch menu at this cafe and return the prices as structured fields."

## Integration

Remote MCP: https://errand.be/api/mcp
REST API: https://errand.be/api/v1
API docs: https://errand.be/docs
Claude plugin: https://github.com/ChloePark85/errand-plugin
Support: support@errand.be