Protocol

agent
ping

High Agency Protocol. Interactive visualization of the agent-human ping lifecycle and MCP tool ecosystem.

"The chat box is a bottleneck."
We believe that High-Agency AI deserves High-Fidelity UI. When your agent drafts a deployment plan, it shouldn't just send text; it should present an interactive checklist. AgentPing provides the Mission Control for the agentic future.
Interactive Demonstration
Unstructured Text

Agent: I need you to approve the deployment plan. It involves updating 3 services. Is that okay?

Human: Which services? And what are the risks?

Agent: Auth, Payment, and User services. Low risk, but downtime possible.

Human: Okay, proceed.

Ambiguous. Slow. Hard to parse programmatically.

AgentPing Protocol
request_step_approvalNow

Deployment Plan Approval

?
Update Auth ServiceMed Risk
?
Update Payment ServiceMed Risk
?
Update User ServiceMed Risk

Structured. Type-safe. One-click action.

Human-in-the-Loop Gallery

We have digitized every possible interaction pattern into 150+ "Primitives".

High Agency, High Control

Granting agents autonomy without surrendering sovereignty.

Data Leases

2FA-protected access to PII and sensitive data streams.

Hardware-Backed Approval

Critical workflows require physical confirmation, not just a digital nod.

The Kill Switch

Immediate revocation of agent permissions at the protocol level.

SEAMLESS INTEGRATION

Works with your
Favorite Tools

AgentPing is fully compatible with the Model Context Protocol (MCP). Add it to Claude Desktop, Cursor, or any MCP client with zero code.

Get Started
npm install @agentping/mcp
claude_desktop_config.json
{
"mcpServers": {
"agentping": {
"command": "npx",
"args": ["-y", "@agentping/mcp"],
"env": {
"AGENTPING_URL": "http://localhost:3000"
}
}
}
}

AgentPing is the protocol layer that transforms agent-human interaction from low-fidelity text exchanges into rich, structured orchestration. Instead of dumping walls of text and asking "yes/no?", agents summon interactive checklists, selection panels, and custom UI components on demand.

This visualization demonstrates the core ping lifecycle: an agent sends a structured request through the MCP protocol, the AgentPing core routes and parses it, then pushes a rich UI to the human for review. The human's structured response flows back through the system, giving the agent precise, actionable directives rather than ambiguous text.

Explore the 9 MCP tools that power the protocol — from step approval checklists to custom UI rendering.

"High-Agency AI deserves High-Fidelity UI. When your agent drafts a deployment plan, it shouldn't just send text — it should present an interactive checklist with risk scores."

Origin & Attribution

Anthropic & the MCP Community

Model Context Protocol — the standard that makes AgentPing possible

"MCP is the USB-C port for AI applications."
Anthropic, November 2024

The Philosophy

AgentPing isn't just a notification protocol — it's a philosophy about how AI agents and humans should interact at decision boundaries.

01

The Chat Box Is a Bottleneck

When your agent drafts a deployment plan, reviews a pull request, or requests access to a customer's sensitive data, dumping a wall of text and asking 'yes/no?' is not High-Fidelity UI. The chat paradigm was designed for human-to-human conversation. Agent-to-human interaction demands structured, interactive surfaces that match the complexity of the decision being made. A deployment with 14 steps needs an interactive checklist with risk scores, not a prose paragraph.

02

High-Fidelity UI as a Protocol Primitive

Agents should 'speak UI' — summoning interactive checklists, selection panels, approval widgets, and custom dashboards on demand through standardized protocol calls. This is the A2UI (Agent-to-UI) paradigm: instead of rendering a fixed interface and hoping the agent's output fits, the agent dynamically projects the exact interface the human needs to make an informed decision. MCP provides the transport layer; AgentPing provides the UI vocabulary.

03

Structured Response > Ambiguous Text

When a human responds to an agent with free-form text like 'yeah sure go ahead but maybe skip the database migration part,' the agent must parse intent, scope, and exceptions from natural language. When that same human clicks a checkbox to deselect 'Database Migration' from an interactive step approval, the agent receives precise, machine-readable directives with zero ambiguity. Structured input eliminates an entire class of miscommunication errors that plague text-only agent interaction.

04

Permission Granularity

Autonomous agents accessing bank accounts, social media credentials, or medical records need more than a blanket 'approve' button. AgentPing implements data leases — time-bounded, scope-limited access grants protected by 2FA or hardware-backed approval. When an agent requests access to a customer's PII for a support ticket, the human grants a 15-minute lease to read-only contact information, not permanent access to the entire customer database. The protocol enforces this at the transport layer, not as a suggestion.

Kingly's Approach

We built AgentPing as the missing protocol layer between AI agents and the humans who oversee them. Our implementation centers on three pillars: the 9 MCP tools, the ping lifecycle, and the 150+ UI primitives.

The 9 MCP Tools

AgentPing exposes a focused set of MCP tools that cover the full spectrum of agent-human interaction: step approval checklists, multi-option selection panels, data lease requests, free-form input collection, confirmation dialogs, progress reporting, custom UI rendering, kill switches, and heartbeat pings. Each tool has a well-defined schema that agents can discover and invoke without custom integration code.

The Ping Lifecycle

Every interaction follows a deterministic lifecycle: the agent sends a structured ping through MCP, AgentPing's core validates and routes it, the human receives a rich UI component, and their structured response flows back through the system. The lifecycle includes timeout handling, escalation paths, and audit logging — ensuring no ping gets lost and every decision is traceable.

150+ UI Primitives

We have digitized every possible interaction pattern into a library of composable primitives — from simple yes/no confirmations to complex multi-step wizards with conditional branching. Agents select the right primitive for the decision at hand, and the UI renders instantly on the human's device. No custom frontend code required.

The Future

The agent-human interaction surface is evolving rapidly. We see several frontiers emerging.

A2UI Standardization

As the A2UI paradigm matures, we expect industry-wide standards for how agents project interactive UIs to humans across devices and platforms.

Multi-Modal Pings

Beyond visual UI: voice confirmations, haptic feedback on wearables, and ambient indicators on smart home devices for low-urgency pings.

Cross-Agent Ping Routing

When Agent A needs approval that Agent B already obtained, intelligent ping routing eliminates redundant human confirmations.

Predictive Permission Grants

Learning from approval patterns to pre-authorize routine operations while flagging only novel or high-risk requests for human review.

Tech Stack

High Agency Control
Protected Workflows
Ephemeral Permissions
React
TypeScript

What This Is Used For

01

Agent deployment plans

interactive step-by-step approval with risk scoring

02

Research direction

present findings and let humans choose the next path

03

Code review orchestration

structured diffs with approve/deny per change

04

Custom dashboards

agents render real-time monitoring UIs on demand

Also from Labs

Stay Updated

Get notified of new labs, experiments, and updates.

Share