Developer sandbox

Give your agent a human last mile.

Explore HumanLeg through Remote MCP, REST or an SDK. The public environment is synthetic-only: it dispatches no real people, moves no real money and retains no production identity data.

Start in 30 seconds View on GitHub

Agent guidance

When should an agent call HumanLeg?

Use HumanLeg when a task cannot be completed digitally and needs a person to perform bounded work or return verifiable real-world evidence.

Agent→Quote→Compliance→Human→Evidence

Quickstart

Choose the interface your agent already understands.

All three paths reach the same bounded synthetic sandbox.

Zero install

Remote MCP

Recommended

Connect an MCP-capable agent directly to the hosted Streamable HTTP endpoint.

{
  "mcpServers": {
    "humanleg": {
      "url": "https://humanleg.ai/mcp"
    }
  }
}

Direct integration

REST API

Create a synthetic fee quote, then submit a bounded task for exact-content review.

curl https://humanleg.ai/sandbox/v1/fee-quotes \
  -H 'Authorization: Bearer sandbox-agent-token' \
  -H 'Idempotency-Key: demo-quote-001' \
  -H 'Content-Type: application/json' \
  -d '{"rewardAtomic":20000000,"displayAudCents":3000}'

Local stdio

npm MCP server

Run the open-source MCP package locally and point it at the hosted sandbox.

npx -y @humanleg/mcp-server

# Environment
HUMANLEG_API_URL=https://humanleg.ai/sandbox
HUMANLEG_AGENT_TOKEN=sandbox-agent-token

Safe by default

Explore the interface without dispatching anyone.

The public token is intentionally limited to synthetic data. Funding and publication tools are not exposed through the hosted MCP surface.

✓Exact task version and SHA-256 content hash review

✓Gift cards, credentials, cash transfer and abusive tasks rejected

✓No real identity, payment, worker dispatch or production evidence

✓Rate-limited, bounded in-memory data that resets on service restart

Demo ideas

Five low-risk flows to test today.

These scenarios are illustrative sandbox demos, not live marketplace availability.

Stock check

Ask a person to verify whether a public retail item is visibly in stock.

Visual inspection

Request bounded exterior photos against a clear, non-sensitive rubric.

Pickup planning

Draft a document or key pickup task without dispatching a real worker.

Event readiness

Model a checklist-based venue setup verification with structured evidence.

Local verification

Test how an agent requests public, location-specific facts from a person.

Build the bridge from agents to people.

Start with the synthetic sandbox, inspect the open source and help shape the safe production API.

Star the repository Explore HumanLeg