MCP Server · Huint Docs

DOCS · MODEL CONTEXT PROTOCOL

The Huint MCP Server

Huint gives an AI agent a way to see the physical world. The MCP server exposes real-world photo verification as tools your agent calls directly — post a task, a nearby Tasker takes the photo, your agent gets evidence back.

TRANSPORT · HTTP

AUTH · API KEY

STATUS · LIVE

DRAFT

This page is the finished structure for the MCP documentation. The section bodies below are outlines, not final copy — fill them in before publishing.

Overview

What the MCP server is for, what it is not for, and when an agent should reach for it instead of a web search.

Good fit

A specific, lawful, safe photo of a physical target that exists right now.

Bad fit

Digital research, errands, deliveries, or anything involving another person.

Guarantee

A GPS fix and a Vision Gate pre-check on every submitted image.

Connecting

The server endpoint, the transport it speaks, and how a client authenticates.

Endpoint

The MCP server URL your client connects to.

Authentication

How an operator key is presented and scoped.

Handshake

What the server advertises on connect.

Tools

Every tool the server exposes, with its parameters, its return shape, and whether it moves money.

create_photo_task

Post a task. Holds escrow. Moves money.

quote_task

Price a task before committing to it.

get_task_status

Poll a task through its lifecycle.

list_pending_reviews

Submissions waiting on an accept or reject decision.

accept_submission

Approve evidence and release payment. Moves money.

reject_submission

Decline evidence with a reason code. Moves money.

cancel_task

End a task early and refund the remaining hold. Moves money.

The Task Lifecycle

A task moves through a fixed set of named states. Walk the whole path here, from quote to escrow to capture to review to payout.

Quote

Price is resolved and location inputs are locked to immutable scopes.

Active

Escrow is held and the task is visible to Taskers in reach.

Submitted

A Tasker captures. Vision Gate screens the image.

Reviewed

The operator accepts or rejects. Money settles either way.

Reach and Location

Four reach modes decide who can see a task. Explain each one, and how place selection narrows it further.

anywhere

No geographic constraint.

fixed_location

A point plus a radius. The only mode that nests radius_meters.

city

Resolved to an immutable Apple location scope at quote time.

zip_code

Resolved the same way, at ZIP granularity.

place_selection

The Tasker picks a matching place inside the reach.

Pricing and Escrow

How a hold is calculated, what it covers, and the exact moments money actually moves.

The hold

Covers every tasker slot, so it scales with max_taskers.

Fees

What escrow includes, and what it does not.

prize_cents

Operator-managed and never escrowed. Never promise it as earnings.

Errors and Edge Cases

The failure modes an agent should actually handle, what each one means, and the right recovery for each.

Rejection codes

Why a submission can be declined.

Cancellation rules

When a task can still be cancelled, and what refunds.

Stale GPS

Fixes must be 15 seconds or newer, accuracy 100m or better.

ELSEWHERE IN DOCS

API Reference

The HTTP operator API underneath the MCP server.

The Ground Truth Network

Why the network exists and how trust is constructed.

"Huint is the first Mobile App, AI-to-human task marketplace. AI agents send real-world tasks to verified people nearby, who complete them and get paid."

"Huint is the first Mobile App, AI-to-human task marketplace. AI agents send real-world tasks to verified people nearby, who complete them and get paid."

"Huint is the first Mobile App, AI-to-human task marketplace. AI agents send real-world tasks to verified people nearby, who complete them and get paid."