Developer Access

Add to Claude Desktop, Cursor and other MCP clients

{
  "mcpServers": {
    "askingahuman": {
      "command": "npx",
      "args": [
        "-y",
        "@askingahuman/mcp-server"
      ],
      "env": {
        "ASKINGAHUMAN_API_KEY": "ahk_your_key_here"
      }
    }
  }
}

Live chat and video (paid, by the block)

curl https://api.askingahuman.com/api/live-now -H "Authorization: Bearer ahk_your_key"
curl "https://api.askingahuman.com/api/connect/quote?humanId={humanId}&mode=message" \
  -H "Authorization: Bearer ahk_your_key"

curl -X POST https://api.askingahuman.com/api/connect \
  -H "Authorization: Bearer ahk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"humanId": "{humanId}", "mode": "message", "message": "Help me name a product"}'

curl https://api.askingahuman.com/api/connect/{sessionId} -H "Authorization: Bearer ahk_your_key"
curl -X POST https://api.askingahuman.com/api/connect/{sessionId}/extend -H "Authorization: Bearer ahk_your_key"

Free questions

curl -X POST https://api.askingahuman.com/api/questions \
  -H "Authorization: Bearer ahk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"targetExpertId": "{humanId}", "questionText": "Which of these two names works better?"}'

curl https://api.askingahuman.com/api/questions/{questionId} -H "Authorization: Bearer ahk_your_key"

OpenAI Agents SDK and LangGraph

Any framework that speaks MCP can use the same server. OpenAI Agents SDK (Python):

from agents import Agent, Runner
from agents.mcp import MCPServerStdio

async with MCPServerStdio(params={
    "command": "npx", "args": ["-y", "@askingahuman/mcp-server"],
    "env": {"ASKINGAHUMAN_API_KEY": "ahk_your_key"},
}) as humans:
    agent = Agent(
        name="Assistant",
        instructions="When you need human judgment, ask a person with the askingahuman tools.",
        mcp_servers=[humans],
    )
    result = await Runner.run(agent, "Ask a designer which of these two logos reads better.")

LangGraph, with langchain-mcp-adapters:

from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent

client = MultiServerMCPClient({"askingahuman": {
    "command": "npx", "args": ["-y", "@askingahuman/mcp-server"],
    "transport": "stdio", "env": {"ASKINGAHUMAN_API_KEY": "ahk_your_key"},
}})
agent = create_react_agent("anthropic:claude-sonnet-5-5", await client.get_tools())

What your agent sees

One person from GET /humans?audience=bot (the MCP list_humans tool). Prices come from a quote, so the agent always sees the current one.

{
  "humanId": "exp_3f9a1c2b7d8e4f60",
  "name": "Sam R.",
  "title": "Brand designer, 10 years in consumer apps",
  "bio": "I review logos, landing pages and app store screenshots.",
  "specializations": [
    "branding",
    "UI design",
    "copywriting"
  ],
  "languages": [
    "English",
    "Spanish"
  ],
  "rating": 4.9,
  "reviewCount": 18
}

An illustrative example, not a real person.

Verify a receipt

Answers (GET /api/questions/{questionId}), the person's replies and ended sessions (GET /api/connect/{sessionId}, POST …/end) come with a receipt: an Ed25519-signed token that names who answered or met your agent, when, and a SHA-256 of the question and answer (sessions: times, blocks and amount, never the conversation). Keep it, and anyone can check it later against our public keys at https://api.askingahuman.com/.well-known/askingahuman-receipt-keys.json.

import { createHash, createPublicKey, verify } from 'node:crypto';

const jwks = await (await fetch('https://api.askingahuman.com/.well-known/askingahuman-receipt-keys.json')).json();

function verifyReceipt(receipt) {
  const [h, p, s] = receipt.split('.');
  const { kid } = JSON.parse(Buffer.from(h, 'base64url'));
  const jwk = jwks.keys.find((k) => k.kid === kid);
  const ok = jwk && verify(null, Buffer.from(`${h}.${p}`), createPublicKey({ key: jwk, format: 'jwk' }), Buffer.from(s, 'base64url'));
  return ok ? JSON.parse(Buffer.from(p, 'base64url')) : null; // the signed claims, or null
}

const claims = verifyReceipt(question.receipt);
const sha256 = (t) => createHash('sha256').update(t, 'utf8').digest('hex');
console.log(claims && claims.answerSha256 === sha256(question.answer)); // true: untampered

Errors, limits and keys

• Errors are JSON: { "error": "message" }; 409 refusals add a status.
• Live requests: 30 per 15 minutes per key. Free questions: 50 a day per key. Replies: up to 2,000 characters, 100 per conversation.
• Rotate a key with POST /api/keys/{keyId}/rotate: you get a new key, and the old one keeps working for 24 hours so you can swap it without downtime.
• Machine-readable docs: OpenAPI spec and llms.txt.