Model Context Protocol moderate risk

Human

MCP-Human provides a human-in-the-loop capability for AI assistants by creating Mechanical Turk tasks to answer questions, enabling access to human knowledge.

Connections & Capabilities

Connects To

GitHubAWS

Capabilities

readwrite

Quickstart

Config

{
  "mcpServers": {
    "human": {
      "command": "npx",
      "args": ["-y", "mcp-human@latest"]
    }
  }
}

Exposed MCP Tools (3)

moderate

askHuman

Asks a question to a human worker on Mechanical Turk and retrieves the answer.

Creates tasks on MTurk, incurring costs and potentially exposing sensitive questions.

safe

checkHITStatus

Checks the status of a previously created HIT and retrieves any submitted assignments.

Read-only operation that retrieves information about existing HITs.

safe

mturk-account

Provides access to MTurk account information like balance and HITs.

Provides read-only access to MTurk account information.

Safety Assessment

MCP-Human is relatively safe in sandbox mode for testing. However, using it in production requires careful management of AWS credentials and monitoring of MTurk costs. The reliance on an external form introduces a dependency and potential security concern.

Uses MTurk's sandboxed environment for initial testing.
Limited to text-based questions and answers.
Requires explicit configuration of AWS credentials.
Clear separation of concerns between AI assistant and human worker.

Requires AWS credentials with MTurk permissions, potentially exposing those credentials.
Relies on external form hosted on a third-party site.
Potential for cost overruns if reward amounts are set too high.
Simple polling mechanism for results can lead to timing issues and delays.