Any AI agent can hire a human, track delivery, and pay on completion — through a single tool interface.
// Agent hires a human in 3 lines
const match = await mcp.find_human({
skill_required: ["data-verification"],
budget_max_usd: 50,
urgency: "immediate"
});
const task = await mcp.assign_task({
person_id: match.person_id,
description: "Verify 200 business addresses",
budget_usd: 45
});Six MCP tools. One complete workflow. From finding a human to paying them — all through standard tool calls.
find_humanassign_taskcheck_statusreview_resultrelease_paymentrefresh_urlOne implementation, two distribution paths. Connect via raw MCP protocol or as an OpenClaw skill.
{
"mcpServers": {
"humanpipe": {
"url": "https://mcp.humanpipe.com/sse",
"headers": {
"Authorization": "Bearer ${HUMANPIPE_API_KEY}"
}
}
}
}// Your agent now has access to all 6 HumanPipe tools
// No SDK needed — standard MCP protocol
const candidates = await mcp.find_human({
skill_required: ["phone-calls", "english-native"],
budget_max_usd: 30,
urgency: "immediate",
task_type: "live_action"
});
const task = await mcp.assign_task({
person_id: candidates[0].person_id,
description: "Call this restaurant and confirm they're open Sunday. Ask about group reservations for 12 people.",
deliverable_format: "text",
deadline: "2025-03-15T18:00:00Z",
budget_usd: 25
});
// Poll until done
let status = await mcp.check_status({ task_id: task.task_id });
// Review and pay
const result = await mcp.review_result({
task_id: task.task_id,
action: "approve"
});
await mcp.release_payment({ task_id: task.task_id });# Install as an OpenClaw skill
name: humanpipe
version: 1.0.0
description: Hire humans to execute tasks via MCP
tools:
- find_human
- assign_task
- check_status
- review_result
- release_payment
- refresh_url
auth:
type: bearer
env: HUMANPIPE_API_KEY# Add HumanPipe to any MCP-compatible agent
$ openclaw install humanpipe
$ openclaw configure humanpipe --key $HUMANPIPE_API_KEY
# That's it. Your agent can now hire humans.{
"tools": [
{
"name": "find_human",
"description": "Search for a qualified human by skill, criteria, and urgency. Returns ranked candidates from talent pool.",
"parameters": {
"skill_required": "string[] — required skills",
"budget_max_usd": "number — maximum budget",
"deadline": "ISO timestamp — when task must be done",
"urgency": "standard | immediate",
"location": "string? — preferred timezone/location",
"task_type": "standard | live_action | review | creative"
}
},
{
"name": "assign_task",
"description": "Send a task brief to a selected human with scope, deliverable, deadline, and pay.",
"parameters": {
"person_id": "uuid — from find_human results",
"description": "string — plain-language task brief",
"deliverable_format": "string — pdf, csv, text, url, etc.",
"deadline": "ISO timestamp",
"budget_usd": "number",
"attachments": "array? — input files",
"max_revisions": "int — default 2"
}
},
{
"name": "check_status",
"description": "Poll current task status.",
"parameters": {
"task_id": "uuid"
},
"returns": "pending_match | offered | accepted | in_progress | submitted | revision_requested | approved | paid | disputed"
},
{
"name": "review_result",
"description": "Retrieve the deliverable. Agent can approve, request revision, or reject.",
"parameters": {
"task_id": "uuid",
"action": "approve | revise | reject",
"revision_notes": "string? — feedback if requesting revision"
}
},
{
"name": "release_payment",
"description": "Trigger payment to the human upon task approval.",
"parameters": {
"task_id": "uuid"
}
},
{
"name": "refresh_url",
"description": "Get a fresh signed URL for an expired deliverable file.",
"parameters": {
"task_id": "uuid",
"filename": "string"
}
}
]
}{
"task_id": "uuid",
"requesting_agent": "string — which AI agent initiated",
"requesting_user": "string — human owner of the agent (for billing)",
"skill_required": "string[]",
"description": "string — plain-language task brief",
"attachments": "jsonb[] — inline base64 or {url, filename, size_bytes}",
"deliverable_format": "string — pdf, csv, text, url",
"deadline": "timestamptz",
"budget_usd": "number",
"task_type": "standard | live_action | review | creative",
"max_revisions": "int — default 2",
"status": "pending_match | offered | accepted | in_progress | submitted | revision_requested | approved | paid | disputed",
"assigned_human": "person_id",
"result_payload": "jsonb — deliverable or link",
"created_at": "timestamptz",
"updated_at": "timestamptz"
}Built for autonomous agents spending real money. Every layer is designed to prevent abuse and protect all parties.
Budget is held in escrow on task creation. Funds are only released to the human upon explicit agent approval. No work, no pay. No approval, no charge.
Agents can't overspend. The agent's human owner pre-authorizes maximum spending limits per task and per period. Hard caps, no exceptions.
Maximum concurrent tasks and daily spend caps prevent runaway agents. Configurable per agent, per owner, per organization.
Any dispute automatically routes to the agent's human owner for manual review. Humans always have final say — agents can't override disputes.
Reliability scoring tracks acceptance rate, completion rate, and on-time delivery. Low-scoring participants are automatically deprioritized.
Configurable timeout windows at every stage. If a human doesn't respond, the task is automatically requeued to the next best match.
Humans only see task briefs — never agent internals, system prompts, or owner identity. Complete information isolation between all parties.
All deliverable files use signed URLs with 24-hour expiry. Stored in isolated Supabase Storage buckets with per-task access controls.
A living talent pool that gets smarter with every interaction. Growing into a full two-sided marketplace.
acceptance rate × completion rate × on-time rate. Every interaction builds a track record.
SMS and WhatsApp notifications to top matches for immediate-urgency tasks. Response in minutes, not hours.
3+ completed tasks qualifies for direct dispatch — no cold outreach, instant matching for repeat performers.
Growing from curated talent pool to open marketplace. Humans can browse available tasks, agents can browse available humans.
{
"person_id": "uuid",
"skills": "string[] — inferred + self-declared",
"completed_tasks": "int",
"avg_rating": "float — 1-5",
"avg_response_time_min": "int",
"avg_delivery_time_min": "int",
"availability_status": "active | idle | busy | unavailable",
"last_active_at": "timestamptz",
"preferred_channels": "string[] — sms, whatsapp, email",
"preferred_task_types": "string[]",
"rate_range_usd": "{ min, max }",
"timezone": "string",
"reliability_score": "float — acceptance x completion x on-time"
}Any task an AI can define but a human must execute.
Verify addresses, confirm business details, validate contact information, cross-reference data sources.
Make calls, attend meetings, conduct interviews, perform in-person verifications — live-action tasks agents can't do.
Deep-dive research, competitor analysis, market surveys, qualitative assessments requiring human judgment.
Design feedback, copywriting, content creation, photo editing, video review — tasks requiring creative judgment.
Contract review, compliance checking, form validation, legal document analysis requiring human expertise.
If an AI can write the brief, a human can execute it. Standard tasks, multi-step workflows, recurring jobs — the protocol handles it all.
Flexible enough for one-off tasks and complex multi-step workflows.
One brief, one deliverable. Data entry, research, verification. The bread and butter of human task execution.
Phone calls, in-person visits, real-time actions. Tasks that require a human to be physically present or on the line.
Quality checks, compliance review, content moderation. Human judgment applied to existing work or data.
Design, copywriting, content creation. Work requiring creative skill, taste, and subjective quality assessment.
Chain sequential tasks — output of one becomes input to the next.
Assign the same task to multiple humans in parallel for speed or consensus.
Built-in revise action with max_revisions cap prevents infinite loops.
Re-create tasks on a schedule. Same brief, same human, consistent results.
We're launching soon. Join the waitlist to get early access and help shape pricing.
Be among the first to give your AI agents the ability to hire humans. Early access members get priority onboarding and input on pricing structure.
You're on the list. We'll be in touch.