mcp-human
Human Assistance for AI Assistants
claude mcp add --transport stdio olalonde-mcp-human npx -y mcp-human@latest \ --env FORM_URL="URL where the form is hosted (HTTPS)" \ --env AWS_REGION="AWS region for MTurk" \ --env AWS_PROFILE="AWS profile to use for credentials" \ --env MTURK_SANDBOX="Use MTurk sandbox (true) or production (false)" \ --env DEFAULT_REWARD="The reward amount in USD"
How to use
This MCP server enables AI assistants to obtain human input by creating tasks (HITs) on Amazon Mechanical Turk. It exposes tools that let an AI ask a human worker a question and later check on the status of submitted responses. Specifically, you can use the askHuman tool to pose a question, specify a reward, and set HIT metadata such as title and validity. If a worker submits an answer within the HIT window, the response is returned to the AI; otherwise you can retrieve the HIT ID and check status later with checkHITStatus. The server relies on the external form hosted for MTurk interactions and uses the MCP protocol to exchange questions and answers between the AI agent and human workers.
How to install
Prerequisites: Node.js 16+ installed, AWS credentials with MTurk permissions, and optionally the AWS CLI for configuring credentials.
-
Install Node.js 16+ from the official site or via your system package manager.
-
Ensure AWS credentials are available (AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY) and configure a profile if desired (e.g., mcp-human) using the AWS CLI:
aws configure --profile mcp-human
-
Install and run the MCP server via npx (no global installation required):
Start in sandbox mode by default
npx -y mcp-human@latest
If you want to run against production MTurk, set MTURK_SANDBOX=false in your environment before starting the server
-
Optional: provide environment variables to customize behavior (examples):
export MTURK_SANDBOX=true export AWS_REGION=us-east-1 export AWS_PROFILE=mcp-human export DEFAULT_REWARD=0.05 export FORM_URL=https://syskall.com/mcp-human/
-
Integrate with your MCP client configuration by adding the server under mcpServers, for example:
{ "mcpServers": { "human": { "command": "npx", "args": ["-y", "mcp-human@latest"] } } }
Additional notes
Tips and caveats: Start with MTurk Sandbox (MTURK_SANDBOX=true) to test end-to-end flows without spending real money. The server uses MTurk's ExternalQuestion format, which requires hosting a form externally (the hosted form URL is configured via FORM_URL). The Default Reward value sets the base HIT payment; adjust as needed, especially when moving from sandbox to production. Ensure your AWS profile (AWS_PROFILE) has AmazonMechanicalTurkFullAccess or appropriate permissions. If you encounter HIT rendering or form-hosting issues, verify that FORM_URL is accessible via HTTPS and that the MTurk form works with the expected query parameters. The tooling supports two main actions: askHuman to create a HIT and checkHITStatus to poll for results. Remember that some constraints are present, such as limiting to simple text-based exchanges and basic polling instead of webhooks.
Related MCP Servers
vibe
🎨 Vibe Design System - Official monday.com UI resources for application development in React.js
time
⏰ Time MCP Server: Giving LLMs Time Awareness Capabilities
rod
Model Context Protocol Server of Rod
browserai
A powerful Model Context Protocol (MCP) server that provides an access to serverless browser for AI agents and apps
AI-web mode
一个基于 MCP (Model Context Protocol) 的智能对话助手Web应用,支持实时聊天、工具调用和对话历史管理。
mcp
🌐 MCP Server for Semantic Search of Macro Data for Quant Research AI Agents