agenticorg-mcp-server

agenticorg-mcp-server

MCP (Model Context Protocol) server for AgenticOrg. It exposes AgenticOrg AI agents, A2A discovery, SOP parsing/deploy, connector discovery, and platform MCP tool discovery to any MCP-compatible client.

Quick Start

npm install -g agenticorg-mcp-server
AGENTICORG_API_KEY=your-key agenticorg-mcp

Or run directly with npx:

AGENTICORG_API_KEY=your-key npx agenticorg-mcp-server

Configure in ChatGPT / Claude Desktop / Cursor

Add to your MCP client configuration:

{
  "mcpServers": {
    "agenticorg": {
      "command": "npx",
      "args": ["agenticorg-mcp-server"],
      "env": {
        "AGENTICORG_API_KEY": "your-api-key"
      }
    }
  }
}

Config file paths by platform

Client	macOS	Windows
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json	%APPDATA%\Claude\claude_desktop_config.json
ChatGPT Desktop	Currently in beta; config path may change. Check OpenAI docs for the latest.	Currently in beta; config path may change.
Cursor	.cursor/mcp.json in your project root	.cursor/mcp.json in your project root

Available Tools

Tool	Description
list_agents	List all AI agents, filter by domain
run_agent	Run any agent, including commerce_sales_agent
get_agent_details	Get full agent config and capabilities
create_agent_from_sop	Parse SOP text and create a draft agent
deploy_agent	Deploy an agent configuration
list_connectors	List native connectors and status for discovery
list_mcp_tools	Discover AgenticOrg agent tools exposed by the platform MCP API
discover_agents_a2a	A2A protocol agent discovery
get_agent_card	Get the public A2A Agent Card

Connector tools are not exposed as direct MCP tools. Use run_agent for the agent that owns the connector workflow, or run list_mcp_tools to see platform agent tools such as agenticorg_commerce_sales_agent.

Authentication

Set one of these environment variables:

• AGENTICORG_API_KEY: API key from your AgenticOrg dashboard (Settings > API Keys)
• AGENTICORG_GRANTEX_TOKEN: Grantex grant token for delegated agent access

Optional:

• AGENTICORG_BASE_URL: custom API base URL (default: https://app.agenticorg.ai)

Example: Run a Commerce Agent from ChatGPT

Once configured, you can say in ChatGPT:

"Use AgenticOrg to show a read-only buyer discovery preview for merchant_demo."

ChatGPT will call the run_agent tool with:

{
  "agent_type": "commerce_sales_agent",
  "action": "buyer_discovery_preview",
  "inputs": {
    "merchant_id": "merchant_demo",
    "request_text": "show a read-only buyer discovery preview"
  }
}

Example: Create an Agent from SOP via ChatGPT

You can create a new agent draft directly from a standard operating procedure:

"Use AgenticOrg to create an agent from this SOP: Step 1 - Receive vendor invoice. Step 2 - Validate GSTIN. Step 3 - 3-way match with PO and GRN. Step 4 - If amount > 5L, escalate to CFO."

ChatGPT will call the create_agent_from_sop tool with:

{
  "sop_text": "Step 1: Receive vendor invoice\nStep 2: Validate GSTIN on GST portal\nStep 3: 3-way match with PO and GRN\nStep 4: If amount > 5L, escalate to CFO",
  "domain_hint": "finance"
}

The server parses the SOP, generates an agent configuration, and returns the draft for review before deployment.

Troubleshooting

"AGENTICORG_API_KEY not set" error

Make sure the env block in your MCP config includes the key, or export it in your shell before running:

export AGENTICORG_API_KEY=your-api-key

"Connection refused" or server not starting

• Verify Node.js >= 18 is installed: node --version
• Try running manually first: AGENTICORG_API_KEY=your-key npx agenticorg-mcp-server
• Check that no firewall or proxy is blocking local stdio communication

"Tool not found" when calling a connector tool

• Direct connector tools are not exposed by this MCP server.
• Use run_agent for the agent that owns the connector workflow.
• Run list_mcp_tools to see agent tools exposed by the platform.

Claude Desktop / Cursor not detecting the server

• Ensure the config JSON is valid.
• Restart the client after editing the config file.
• Check the client's MCP log for error messages.

License

Apache-2.0