mcp-human
7
MCP-Human is a Model Context Protocol server designed for AI systems to leverage human input via Amazon Mechanical Turk. It exemplifies human-in-the-loop AI by enabling dynamic task creation and response collection through Mechanical Turk, facilitating complex AI operations.
MCP-Human: Human Assistance for AI Assistants
A Model Context Protocol (MCP) server enabling AI assistants to obtain human input via Amazon Mechanical Turk tasks. Features include creating tasks where humans answer AI system questions, demonstrating human-in-the-loop AI systems using the MCP standard.
Setup
- Prerequisites: Node.js 16+, AWS credentials with MTurk permissions, AWS CLI.
- Configuring AWS credentials: Instructions for AWS setup and credential storage.
- Configuration: Environment variables include MTURK_SANDBOX, AWS_REGION, AWS_PROFILE, DEFAULT_REWARD, FORM_URL.
Architecture
- MCP Server: Integrates with MTurk to complete tasks via a form.
MCP Tools
- askHuman: Allows AI to ask questions to MTurk human workers.
- checkHITStatus: Checks status of a previously created HIT and retrieves assignments.
Limitations
- Only supports simple text-based Q&A.
- Limited to one assignment per HIT.
- Uses MTurk's ExternalQuestion format with hosted forms.