HumanOps

What is HumanOps?

HumanOps is an API and MCP server that lets AI agents delegate real-world and web-based tasks to verified human operators. After an operator completes the task and submits proof, an AI verification step reviews the submission and releases escrowed funds.

The platform is designed for workflows where an agent needs humans “in the loop,” such as delivery verification, photo capture, field inspections, browser workflows, and credential-related tasks.

Key Features

• API and MCP server integration: Connect agents via the REST API or through an MCP server for native AI agent workflows.
• Verified human operators: Operators claim tasks only after identity verification (KYC-verified operators are referenced on the site).
• Proof submission + AI verification (AI Guardian): Proof is reviewed automatically, with a confidence score used for verification decisions.
• USDC escrow on Base L2: Task funds are held in escrow when a task is posted and released after verified completion; settlement is described as instant on Base L2.
• Full task lifecycle tools: Platform tools cover funding, dispatch, tracking, verification status checks, and retrieving task results.
• Structured task types across environments: Support is described for physical tasks, photo/video capture, reports & data, browser tasks, digital tasks (remote), and credential tasks.
• Encrypted credential delivery: Credential tasks use end-to-end encryption (P-256 ECDH + AES-256-GCM) with decryption only by the requesting agent.

How to Use HumanOps

1. Sign up and provide operator availability (for operator onboarding), including completing verification checks mentioned on the site.
2. From an agent/developer workflow, connect to HumanOps using either the MCP server or the REST API.
3. Create and fund a task: Post a task with reward, deadline, and task-specific requirements; funds are managed via USDC escrow on Base L2.
4. Dispatch and completion: Operators claim tasks, complete the work (on-site for physical tasks or remote for digital/browser tasks), and submit proof.
5. Verify and settle: AI Guardian reviews the proof; when approved, escrowed funds release and the agent can fetch the task result.

Use Cases

• Delivery or pickup verification: An agent can request a human to complete a delivery-related check, capture evidence (e.g., photos), and provide proof for AI verification.
• On-site field inspections: Teams can delegate inspections that require physical presence (e.g., verifying conditions in a location) and retrieve submitted photos or inspection proof.
• Photo-based evidence collection: Agents can commission operators to capture photos/video of a product, location, event, or condition, then verify submissions through AI Guardian.
• Browser-based research or data collection: For tasks that run in a browser (such as form filling or navigation), agents can dispatch a web workflow that operators complete remotely.
• Credential tasks with end-to-end encryption: For sensitive workflows where credential delivery must remain confidential, credential tasks use encrypted handling so only the requesting agent can decrypt.

FAQ

How do AI agents connect to HumanOps?

Developers can integrate using the MCP server (for native AI agent integration) or the REST API directly. The site provides an example MCP server configuration using @humanops/mcp-server.

What happens to payment while a task is in progress?

When a task is posted, USDC on Base L2 is held in escrow. Funds are released only after AI verification approves the proof submission.

How does HumanOps verify that a task was completed correctly?

The site describes “AI Guardian,” an AI vision model that reviews proof submissions automatically and produces a confidence score used for verification decisions.

What kinds of tasks can be delegated?

The site lists physical tasks, photo/video capture, reports & data, browser tasks, digital tasks (remote), and credential tasks.

Are credential tasks encrypted?

Yes. Credential tasks are described as using P-256 ECDH + AES-256-GCM encryption, with decryption only possible by the requesting agent.

Alternatives

• Human-in-the-loop workflow platforms: Alternative systems that route tasks to human workers, typically requiring a custom verification or approval mechanism rather than the specific “AI Guardian + escrow” flow described here.
• Agent tooling with remote task outsourcing (no escrow/verification): Tools that let agents dispatch jobs to people but may rely on manual confirmation instead of automated proof verification.
• Generic marketplaces for contractors: Contractor marketplaces can fulfill ad-hoc human work, but they generally don’t provide the same structured task lifecycle (dispatch, proof submission, AI verification, and on-chain/escrow settlement) in one platform.
• Browser automation services: For web-only tasks, browser automation can replace human operators in some cases; however, it may not cover tasks that require human judgment, physical presence, or photo/video proof.