Introducing ?? Agentic MCP: An OpenAI Agents API MCP Server
Using the new Agentics MCP for OpenAi Agents Service, I deployed 500 agents, at once. Not hypothetical, real agents, in production, executing tasks at scale. This is what’s possible now with the Agentic MCP NPM.
The core idea is simple: kick off agents, let them run, and manage them from your chat or code client like Cline, Cursor, Claude, or any service that supports MCP. No clunky interfaces, no bottlenecks, just pure autonomous orchestration.
Need a research agent to search the web? Spin one up, that agent can then spawn sub agents and those can also. Need agents that summarize, fetch data, interactively surf websites, or interact with customers? Done. This isn’t about AI assistants anymore; it’s about fully autonomous agent networks that execute complex workflows in real time.
This system is built on OpenAI’s Agents API/SDK, using TypeScript for flexibility and precision. The MCP architecture allows agents to coordinate, share context, and escalate tasks without human micromanagement. And now, thanks to the Agentics Foundation, it’s available as an NPM package—open and ready to deploy.
I’ve been testing this in real-world scenarios, and the impact is immediate. Scaling automation isn’t just possible—it’s trivial. The ability to launch specialized agents, let them self-organize, and achieve goals autonomously changes everything. Whether for research, customer support, or structured decision-making, this is how AI should operate.
Thank you to everyone supporting this project.
The future of agent-based automation is here. Let’s build.
?? Now live as an NPM package: npm install -g @agentics.org/agentic-mcp
??Paid for by the Agentics Foundation. (https://agentics.org)
What is it?
A powerful Model Context Protocol server with advanced AI capabilities by the Agentics Foundation. Built on the OpenAI Agents API/SDK using TypeScript, this package implements a comprehensive MCP server that enhances AI agents with sophisticated tools and orchestration capabilities:
?? Core Capabilities
?? Installation
# Install globally
npm install -g @agentics.org/agentic-mcp
# Or as a project dependency
npm install @agentics.org/agentic-mcp
??? Architecture
The Agentic MCP server is built on a modular architecture that enables seamless integration between AI agents and external tools:
Core Components
Agent Types
?? Configuration
Create a configuration file for the MCP server. Here's a sample configuration:
{
"mcpServers": {
"openai-agent": {
"command": "node",
"args": [
"dist/index.js"
],
"env": {
"OPENAI_API_KEY": "YOUR_API_KEY_HERE",
"SUPABASE_URL": "https://your-supabase-project.supabase.co",
"SUPABASE_KEY": "YOUR_SUPABASE_KEY_HERE",
"LLM_DEBUG": "true",
"AGENT_LIFECYCLE": "true",
"TOOL_DEBUG": "true"
},
"disabled": false,
"autoApprove": [
"research",
"support",
"customer_support",
"database_query",
"handoff_to_agent",
"summarize"
]
}
}
}
??♂? Usage
As a Command Line Tool
# Set required environment variables
export OPENAI_API_KEY=your_api_key_here
export SUPABASE_PROJECT_ID=your_project_id
export SUPABASE_ACCESS_TOKEN=your_access_token
# Run the MCP server
agentic-mcp
As a Library
import { OpenAIAgentMCPServer } from '@agentics.org/agentic-mcp';
const server = new OpenAIAgentMCPServer({
name: 'openai-agent',
version: '1.0.0',
openai: {
apiKey: process.env.OPENAI_API_KEY,
defaultModel: 'gpt-4o-mini'
},
tracing: {
enabled: true,
level: 'debug'
},
tools: {
enabled: ['research', 'database_query', 'customer_support', 'handoff_to_agent', 'summarize'],
config: {
database: {
projectId: process.env.SUPABASE_PROJECT_ID,
key: process.env.SUPABASE_ACCESS_TOKEN
},
openai: {
apiKey: process.env.OPENAI_API_KEY
}
}
},
guardrails: {
enabled: true,
rules: []
}
});
server.serve().catch(error => {
console.error("? Server error:", error);
process.exit(1);
});
?? Advanced Features
Context Management System
The Agentic MCP includes a sophisticated context management system that enables:
// Example of context management
const context = new Context();
context.initializeWorkflow();
context.remember('user_preference', { theme: 'dark' });
context.trackAction('research_initiated');
Agent Orchestration
The system supports sophisticated agent orchestration patterns:
// Example of agent handoff
const handoffTool = {
name: "handoff_to_agent",
description: "Transfer the conversation to another specialized agent",
parameters: {
type: "object",
properties: {
agent_name: {
type: "string",
enum: ["researcher", "database_expert", "customer_support"]
},
reason: {
type: "string"
}
},
required: ["agent_name", "reason"]
},
execute: async (params) => {
// Handoff logic
}
};
Guardrails System
The Agentic MCP includes a configurable guardrails system for ensuring safe and appropriate responses:
// Example guardrail implementation
const customGuardrail = {
async check(msgs, context) {
// Validation logic
return true;
},
onFailure(msgs, context) {
// Failure handling
}
};
Streaming Support
The system supports real-time streaming responses for interactive applications:
// Example of streaming usage
const streamIterator = AgentRunner.run_streamed(agent, [input]);
for await (const event of streamIterator) {
// Process streaming event
console.log(event.delta);
}
??? Creating and Using Tools
Creating a New Tool
// Example tool implementation
export class CustomTool implements MCPTool {
name = 'custom_tool';
description = 'Description of your custom tool';
inputSchema = {
type: 'object',
properties: {
param1: {
type: 'string',
description: 'Description of parameter 1'
}
},
required: ['param1']
};
async execute(params: any, context: Context): Promise<any> {
// Tool implementation
return { result: 'Success' };
}
}
Using the Tools
?? Troubleshooting and Debugging
?? License
MIT
?? Contributors
Created by the Agentics Foundation
Retail & Omni-Channel Specialist | 35+ Years in Retail Ops Management | Achieved 30% Sales Growth via Shopify Integration | Reduced Operational Costs by 15% | Improved Customer Satisfaction by 25%
1 天前500 AI agents. Running. Scaling. Self-organising. Right now. This is not automation. It is autonomy. No bottlenecks. No micromanagement. No excuses. Just AI operating at relentless scale, running workflows faster than any human ever could. The old rules are burned to the ground. Adapt or watch from the sidelines.
Executive Director Design | Design Leadership | Customer experience design and measurement
4 天前Laks Lakshmanan Lakshmanan
Co-Founder and CEO @ Konverso | AI Agent platform
5 天前Alban Petit, PhD
AI since 2012 | Wall St & SV 13 yrs | Techies, Quants | Investor, Builder, Disruptor | Roosetta Private Talent Database 190K, LinkedIn 30K | GTM, Leaders, Csuite, Boards | 193 hires previous 10 yrs | 17 Unicorns 5 IPOs
1 周Love at First Sight ??????????
Operations Research and Development | Geopolitical & Geoeconomics Expert | Veteran Advocate | PhD Candidate | Project Manager | SQL Python R | Ai & Palantir Enthusiast | Active Security Clearance
1 周Excellent work Reuven