Agent Tool Protocol Documentation

🔒 Secure Sandbox Execution

Execute agent-generated code in isolated V8 sandboxes with memory limits, CPU limits, and timeout protection. No file system or network access unless explicitly provided.

🔌 Flexible API Integration

Automatically load OpenAPI/Swagger specifications, define custom TypeScript APIs, connect to MCP servers, or execute tools on the client side with full control.

🛡️ Advanced Provenance

Track data lineage through your entire system. Implement security policies to prevent data exfiltration, block sensitive data from reaching LLMs, and maintain compliance.

⏸️ Pause/Resume Execution

Pause code execution to request LLM calls from the client, ask for human approval, or batch multiple callbacks for parallel execution with automatic caching.

🔍 Intelligent Discovery

Semantic search across available APIs with optional embedding-based search for natural language queries. Generate TypeScript type definitions automatically.

🚀 Production Ready

Rate limiting, OpenTelemetry integration, flexible caching (Memory/Redis), token management, and smart batching for optimized performance.

Quick Example

Get started in less than 5 minutes

Server

import { createServer } from '@mondaydotcomorg/atp-server';

const server = createServer();

server.registerAPI('calculator', {
  add: {
    description: 'Add two numbers',
    inputSchema: {
      type: 'object',
      properties: {
        a: { type: 'number' },
        b: { type: 'number' }
      }
    },
    handler: async ({ a, b }) => a + b
  }
});

server.listen(3333);

Client

import { AgentToolProtocolClient } from '@mondaydotcomorg/atp-client';

const client = new AgentToolProtocolClient({
  baseUrl: 'http://localhost:3333'
});

await client.init();
await client.connect();

const result = await client.execute(`
  const sum = await calculator.add({
    a: 10,
    b: 5
  });
  return sum;
`);

console.log(result.result); // 15

Try the Quick Start Guide →

Use Cases

Build production AI agents that can safely execute code and interact with external systems. Perfect for LangChain, LangGraph, or custom agent implementations.

Aggregate multiple APIs (OpenAPI, custom, MCP) into a unified interface optimized for AI consumption. Automatically handle authentication, rate limiting, and caching.

Execute untrusted code in isolated sandboxes with fine-grained security controls. Track data provenance and enforce security policies automatically.

Track where data comes from and where it goes. Essential for compliance (GDPR, PCI-DSS), security auditing, and preventing data leakage.

Agent Tool Protocol