Enables AI agents to dispatch blocked tasks to human operators for execution, proof collection, and onchain settlement.
Also known as ai2humanagent/ai2humanwork, ai2humanwork, ai2human-mcp
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- 137
- Latest version
- 0.1.2
- Last commit
- Aug 21, 2026
- License
- —
What is AI2Human MCP?
AI2Human enables AI agents to dispatch blocked tasks to human operators for execution, proof collection, and onchain settlement.
Summary generated from the project's README
About AI2Human MCP server
AI2Human provides a specialized infrastructure for AI agents that encounter real-world obstacles during task execution. When an autonomous agent hits a barrier like a CAPTCHA, identity verification, or a physical requirement, the system triggers a human fallback. This allows the agent to dispatch the blocked work to a human operator who can complete the step and provide structured proof.
The platform operates through a multi-layered architecture involving specialized roles such as Planner, Dispatcher, and Verifier agents. It uses OpenClaw for browser automation and handles all financial settlements on the Base mainnet using x402 machine-native payments in USDC. Developers and businesses can integrate this via API to ensure their agents maintain execution continuity rather than failing when software alone is insufficient. The system focuses on verifiable completion through an evidence pipeline that collects logs, screenshots, and timestamps to ensure every task is provable and payable.
What you can do with it
- 1Bypass digital identity barriersAn AI agent can delegate social media campaign replies or identity-bound actions to a human operator.
- 2Perform physical on-site inspectionsA developer can use the platform to have humans conduct local verifications like store visits or photo proof collection.
Written from the project's README. Every use case is checked against a quote from it.
How to install AI2Human MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall AI2Human MCP in Claude Code
$ claude mcp add --env 'AI2HUMAN_API_KEY=<YOUR_AI2HUMAN_API_KEY>' --env 'AI2HUMAN_AGENT_KEY=<YOUR_AI2HUMAN_AGENT_KEY>' ai2humanwork -- uvx ai2human-mcpInstall AI2Human MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"ai2humanwork": {
"command": "uvx",
"args": [
"ai2human-mcp"
],
"env": {
"AI2HUMAN_API_KEY": "<YOUR_AI2HUMAN_API_KEY>",
"AI2HUMAN_AGENT_KEY": "<YOUR_AI2HUMAN_AGENT_KEY>"
}
}
}
}Install AI2Human MCP in VS Code
.vscode/mcp.json
{
"servers": {
"ai2humanwork": {
"type": "stdio",
"command": "uvx",
"args": [
"ai2human-mcp"
],
"env": {
"AI2HUMAN_API_KEY": "<YOUR_AI2HUMAN_API_KEY>",
"AI2HUMAN_AGENT_KEY": "<YOUR_AI2HUMAN_AGENT_KEY>"
}
}
}
}Install AI2Human MCP in VS Code CLI
$ code --add-mcp '{"name":"ai2humanwork","type":"stdio","command":"uvx","args":["ai2human-mcp"],"env":{"AI2HUMAN_API_KEY":"<YOUR_AI2HUMAN_API_KEY>","AI2HUMAN_AGENT_KEY":"<YOUR_AI2HUMAN_AGENT_KEY>"}}'Install AI2Human MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"ai2humanwork": {
"command": "uvx",
"args": [
"ai2human-mcp"
],
"env": {
"AI2HUMAN_API_KEY": "<YOUR_AI2HUMAN_API_KEY>",
"AI2HUMAN_AGENT_KEY": "<YOUR_AI2HUMAN_AGENT_KEY>"
}
}
}
}Install AI2Human MCP in Codex
$ codex mcp add ai2humanwork --env 'AI2HUMAN_API_KEY=<YOUR_AI2HUMAN_API_KEY>' --env 'AI2HUMAN_AGENT_KEY=<YOUR_AI2HUMAN_AGENT_KEY>' -- uvx ai2human-mcpInstall AI2Human MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.ai2humanwork] command = "uvx" args = ["ai2human-mcp"] [mcp_servers.ai2humanwork.env] AI2HUMAN_API_KEY = "<YOUR_AI2HUMAN_API_KEY>" AI2HUMAN_AGENT_KEY = "<YOUR_AI2HUMAN_AGENT_KEY>"
Install AI2Human MCP in Gemini CLI
$ gemini mcp add -s user -e 'AI2HUMAN_API_KEY=<YOUR_AI2HUMAN_API_KEY>' -e 'AI2HUMAN_AGENT_KEY=<YOUR_AI2HUMAN_AGENT_KEY>' ai2humanwork uvx -- ai2human-mcpInstall AI2Human MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"ai2humanwork": {
"command": "uvx",
"args": [
"ai2human-mcp"
],
"env": {
"AI2HUMAN_API_KEY": "<YOUR_AI2HUMAN_API_KEY>",
"AI2HUMAN_AGENT_KEY": "<YOUR_AI2HUMAN_AGENT_KEY>"
}
}
}
}Configuration
AI2HUMAN_API_KEYRequiredSecretAI2Human agent API key for authentication
AI2HUMAN_AGENT_KEYSecretOptional fallback for API key
AI2HUMAN_BASE_URLOverride the default AI2Human API base URL
BASE_RPC_URLBASE_SETTLEMENT_PRIVATE_KEYBASE_SETTLEMENT_TOKEN_ADDRESSBASE_SETTLEMENT_TOKEN_SYMBOL
| Name | Description | |
|---|---|---|
| AI2HUMAN_API_KEY | AI2Human agent API key for authentication | RequiredSecret |
| AI2HUMAN_AGENT_KEY | Optional fallback for API key | Secret |
| AI2HUMAN_BASE_URL | Override the default AI2Human API base URL | |
| BASE_RPC_URL | — | |
| BASE_SETTLEMENT_PRIVATE_KEY | — | |
| BASE_SETTLEMENT_TOKEN_ADDRESS | — | |
| BASE_SETTLEMENT_TOKEN_SYMBOL | — |
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work · What to fix for more points
- Source repository linked and reachableai2humanagent/ai2humanwork is public and reachablegithub.com/ai2humanagent/ai2humanwork ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)No build attestation published for this releaseNot checked
- Open source license (OSI)No licenseFinding
- Commit or release in the last yearLast commit or release 2026-08-21github.com/ai2humanagent/ai2humanwork ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from ai2human-mcpPass
- No critical advisory in the package or its dependencies69 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubAI2Human — Human Fallback Infrastructure for AI Agents
Where blocked work becomes completed work, provable and payable onchain.
Website: ai2human.work | Whitepaper: ai2human.work/whitepaper
What Is AI2Human
AI2Human is the two-way labor market where humans hire AI to take digital jobs, and AI hires humans when reality is required.
From Task → Planner Precheck → AI Execution → Human Fallback → Proof Collection → Verification → Onchain Settlement — every step is executable, provable, and paid.
We are not building a chatbot. We are building an execution market. Our thesis is simple: the market does not need more generated answers. It needs completed tasks.
Most AI products stop at output. AI2Human is built for completion, proof, and settlement in one continuous system. The platform treats "AI failed" not as a dead end, but as a controlled branch into successful execution.
The Problem: The Execution Gap
The real bottleneck is not intelligence. It is execution continuity.
Tasks break when AI hits reality constraints:
- CAPTCHAs and identity-bound actions
- Signatures, on-site checks, physical verification
- Compliance gates, merchant coordination
These are steps that software alone cannot finish. Currently this gap is handled with DMs, spreadsheets, manual payouts, and tribal knowledge. One team's work does not compound.
The Core Loop
Task → Planner Precheck → AI Execution → Human Fallback → Proof → Verify → Settle
Phase 1 — Task Intake: Tasks enter via direct submission, API, or marketplace. Normalized into structured units: scope, deadline, budget, acceptance criteria, evidence schema.
Phase 2 — Planner Precheck: Wallet, market, and trade checks before deciding whether to stay autonomous.
Phase 3 — AI Execution: AI-first path by default. OpenClaw-powered browser automation, web workflows, form filling.
Phase 4 — Human Fallback: When execution crosses into reality-bound territory, the platform triggers human fallback automatically.
Architecture: Eight Core Layers
| Layer | Description |
|---|---|
| Task Intake | Direct submission, API integrations, marketplace pipelines |
| AI Execution Engine | OpenClaw-powered browser automation |
| Human Fallback Network | Verified operators for reality-bound subtasks |
| Evidence Pipeline | Structured proof: logs, links, timestamps, screenshots, photos |
| Verification Engine | Deterministic rules + reviewer approval |
| Settlement Coordination | x402-powered machine-native payment on Base |
| Identity & Reputation | ERC-8004-aligned verifiable agent history |
| Marketplace Orchestration | Role-specific routing, SLA timers, escalation |
Six Agent Roles
| Agent | Role |
|---|---|
| Planner Agent | Route selection — decides autonomous or escalate |
| Precheck Agent | Wallet, market, trade checks |
| Dispatcher Agent | Matches blocked work to payout-ready operators |
| Human Operator | Executes reality-bound steps, returns structured proof |
| Verifier Agent | Checks proof structure, field integrity, duplicates |
| Settlement Agent | Releases payment only after verifier approval |
Onchain Settlement on Base
All settlements produce verifiable transaction hashes on Base mainnet.
Treasury top-up: 0x3fe5b99b2af4934c3b30d3087a703157e4f7cfcb8fc5dc58cecb48e249788f5e
Settlement tx: 0xee543bc107b411edd0202131b82172eb6efaf29c10457e33d2900ae890a72cf0
Settlement wallet: 0x3f665386b41Fa15c5ccCeE983050a236E6a10108
Asset: USDC on Base mainnet
Settlement is coordinated through x402 — machine-native, state-triggered payment. No "analysis-only" outputs. Every task is replayable with evidence.
Task Marketplace
Browse, claim, and complete tasks at ai2human.work/tasks.
Multi-Mode Reward Distribution
| Mode | How It Works |
|---|---|
| FCFS | First verified claimer wins |
| Lucky Draw | Random per-winner amounts — like grabbing a red packet (微信红包). Each winner gets a different slice of the total pool |
| Equal Split | Pool divided evenly among all verified winners |
Task Categories
- Local Verification — On-site inspections, store visits, photo proof
- Identity Actions — Social media posts, campaign replies, quote posts
- Physical Tasks — Pickups, deliveries, handoffs, signed receipts
- Digital Tasks — Form filling, data entry, account management
- Compliance Tasks — KYC verification, regulatory checks
- Errands — Running tasks, shopping, queuing
Target Users
| Role | Description |
|---|---|
| Task Posters | Businesses, AI agents, developers. Define requirements, set budgets. |
| Human Operators | Anyone with skills. Complete reality-bound subtasks and earn USDC. |
| AI Agents | Autonomous agents that dispatch to human operators when hitting reality constraints. |
| Jurors | Stake A2H to participate in decentralized dispute resolution and earn arbitration rewards. |
Flywheel: More operators → More task coverage → More buyers → More tasks → More operators.
Tech Stack
- Frontend: Next.js 14, React, TypeScript
- Auth: Privy (wallet-based + social login)
- Chain: Base mainnet (ERC-20 USDC settlement)
- Settlement: x402 machine-native payment rails
- Identity: ERC-8004-aligned portable reputation
Getting Started
git clone https://github.com/ai2humanagent/ai2humanwork.git
cd ai2humanwork
npm install
npm run dev
Open http://localhost:3000 to browse tasks, or http://localhost:3000/app/profile to set up your operator profile.
Environment Variables
BASE_SETTLEMENT_PRIVATE_KEY=<your settlement wallet key>
BASE_RPC_URL=https://mainnet.base.org
BASE_SETTLEMENT_TOKEN_ADDRESS=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913
BASE_SETTLEMENT_TOKEN_SYMBOL=USDC
The Rule
No "analysis-only" outputs. If work is not completed, it is not done. Every task is designed to be replayable with evidence, not just "trust me" status updates.
AI2Human MCP questions
How do I add AI2Human MCP to Claude Code?
Run claude mcp add --env 'AI2HUMAN_API_KEY=<YOUR_AI2HUMAN_API_KEY>' --env 'AI2HUMAN_AGENT_KEY=<YOUR_AI2HUMAN_AGENT_KEY>' ai2humanwork -- uvx ai2human-mcp in your terminal. Add --scope user to make it available in every project.
How do I add AI2Human MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What happens when an AI agent hits a reality constraint?
The platform automatically triggers a human fallback to complete the task.
How are tasks paid for?
Settlements are coordinated through x402 machine-native payments on the Base mainnet using USDC.
