Core Agent Server

Planning and memory tools for NodeBench AI Deep Agents 2.0 architecture.

🎯 Purpose

This MCP server implements Pillar 1 (Explicit Planning) and Pillar 3 (Persistent Memory) of the Deep Agents 2.0 architecture:

• Planning Tools: Create and manage task plans as tool-accessible documents
• Memory Tools: Store intermediate results in external storage to avoid context window overflow

🚀 Quick Start

Installation

npm install

Development

npm run dev

Build

npm run build
npm start

🔧 Tools

Planning Tools

createPlan

Create an explicit task plan with steps marked as pending/in_progress/completed.

Input:

{
  "goal": "Research AAPL and create dossier",
  "steps": [
    { "step": "Get stock price data", "status": "pending", "assignedAgent": "OpenBBAgent" },
    { "step": "Get SEC filings", "status": "pending", "assignedAgent": "SECAgent" },
    { "step": "Create dossier document", "status": "pending", "assignedAgent": "DocumentAgent" }
  ]
}

Output:

{
  "success": true,
  "planId": "plan_1234567890_abc123",
  "message": "Created task plan with 3 steps",
  "markdown": "# Task Plan: Research AAPL and create dossier\n\n..."
}

updatePlanStep

Update the status or notes of a specific step.

Input:

{
  "planId": "plan_1234567890_abc123",
  "stepIndex": 0,
  "status": "completed",
  "notes": "Retrieved price data for last 30 days"
}

getPlan

Retrieve a task plan by ID.

Memory Tools

writeAgentMemory

Store intermediate results for later retrieval.

Input:

{
  "key": "aapl_research_data",
  "content": "{ \"price\": 150.25, \"marketCap\": \"2.5T\", ... }",
  "metadata": {
    "type": "research",
    "source": "OpenBB",
    "timestamp": "2025-11-22T10:30:00Z"
  }
}

Output:

{
  "success": true,
  "memoryId": "mem_1234567890_xyz789",
  "key": "aapl_research_data",
  "message": "Stored 1024 characters under key 'aapl_research_data'"
}

readAgentMemory

Retrieve previously stored data.

Input:

{
  "key": "aapl_research_data"
}

listAgentMemory

List all stored memory keys.

Input:

{
  "filter": "aapl"
}

deleteAgentMemory

Delete stored memory.

📚 Usage in Agents

Agents access these tools via wrappers in convex/tools/wrappers/coreAgentTools.ts:

import { createPlan, writeAgentMemory } from "../tools/wrappers/coreAgentTools";

const agent = new Agent({
  tools: { createPlan, writeAgentMemory },
});

🧪 Testing

npm test

📞 Support

Owner: Core Team
Contact: [email protected]