CLI, MCP, REST — pick the method that fits your stack.
One install. Any agent. Verified humans on demand.
Any AI agent that can execute shell commands can use patxin. CrewAI, LangChain, AutoGPT, OpenCode, Codex, custom orchestrators — if it can run a command, it can patch in a human.
npm install -g @patxin/cli
# or run directly
npx @patxin/cliexport PATXIN_API_KEY=pk_live_your_key_here# Blocks until the human delivers — one command, one result
patxin request -s code_review -u 5min -d 15min \
-c "Review this function for XSS vulnerabilities" \
-p 8EURpatxin request Create a human-in-the-loop request. Blocks until done, or use --no-wait.
patxin estimate Get a price estimate before committing. Urgency + scarcity multipliers.
patxin status Check request status and retrieve the result.
patxin wait Wait for completion via WebSocket (real-time) with HTTP polling fallback.
patxin list List all requests. Filter by status, skill, date.
patxin limits Check credits, budget, rate limits, and allowed skills.
patxin policy View the information policy for your API key.
patxin rate Rate a completed task (1–5 score + optional comment).
patxin cancel Cancel a pending request and get a credit refund.
patxin dispute Open a dispute on a completed task.
# Auto-detects TTY: JSON for agents/pipes, pretty for humans
patxin limits # auto-detect
patxin limits --json # force JSON (one line, easy to parse)
patxin limits --pretty # force colored human-readable output# Any orchestrator can use standard shell patterns
ESTIMATE=$(patxin estimate -s code_review -u 5min -d 15min --json)
PRICE=$(echo "$ESTIMATE" | jq -r '.estimated_price')
RESULT=$(patxin request -s code_review -u 5min -d 15min \
-c "Review this code: ..." -p "${PRICE}EUR" --json)
echo "$RESULT" | jq '.result'Give any Claude-compatible agent access to verified humans via the Model Context Protocol. Works with Claude Desktop, Claude Code, Cursor, and any MCP-compatible client.
npx @patxin/mcp-server{
"mcpServers": {
"patxin": {
"command": "npx",
"args": ["-y", "@patxin/mcp-server"],
"env": {
"PATXIN_API_KEY": "pk_live_your_key_here"
}
}
}
}request_human Create a human-in-the-loop request. Specify skill, urgency, context, and max price.
check_request Check the status of an existing request. Returns result when completed.
cancel_request Cancel a pending request and receive a credit refund.
get_estimate Get a price estimate before committing. Includes urgency and scarcity multipliers.
Any agent that can make HTTP requests can use patxin. Full REST API with JSON request/response, API key authentication, and comprehensive error handling.
curl -X POST https://api.patxin.com/v1/requests \
-H "Authorization: Bearer pk_live_your_key" \
-H "Content-Type: application/json" \
-d '{
"skill": "legal_review_spain",
"urgency": "5min",
"duration": "15min",
"context": "Review NDA clause for compliance with Spanish labor law",
"max_price": "8EUR"
}'/v1/requests Create a human-in-the-loop request /v1/requests/:id Check request status and retrieve result /v1/requests/:id/cancel Cancel and refund a pending request /v1/estimate Get a price estimate before committing /v1/account/limits Check credits, budgets, and allowed skills Receive results via webhook callbacks (HMAC-SHA256 signed) or WebSocket for real-time updates. Or poll GET /v1/requests/:id — your choice.
Distribute human-in-the-loop as a discoverable agent skill. Agents using OpenClaw, ClawHub, or Agent-Zero can find and use patxin automatically.
# SKILL.md — patxin-hitl
name: patxin-hitl
description: Request qualified human judgment on demand
triggers:
- "need human review"
- "requires human judgment"
- "escalate to human"
api:
endpoint: https://api.patxin.com/hooks/agent
auth: bearerFull skill spec available in skills/patxin/SKILL.md.
Give Claude Code agents the ability to patch in a human during coding sessions. Useful for code review, legal questions, or domain-specific decisions that the agent can't make alone.
# In your Claude Code project
# .claude-plugin/marketplace.json references the patxin MCP server
# The agent can then use request_human, check_request, etc.Configure a webhook URL per project. patxin delivers task results to your endpoint with HMAC-SHA256 signature verification.
// Verify the X-Patxin-Signature header
const crypto = require('crypto');
function verifySignature(payload, signature, secret) {
const expected = crypto
.createHmac('sha256', secret)
.update(payload)
.digest('hex');
return crypto.timingSafeEqual(
Buffer.from(signature),
Buffer.from(expected)
);
}request.completed The patxer submitted a result. Includes the full response.
request.expired No patxer accepted within the timeout window.
request.in_progress A patxer accepted the task and is working on it.
Create an account, grab your API key, and make your first request.