NexusChat MCP plugins

Incorta MCP Plugin

Version: 1.6.1

An Incorta Copilot plugin that integrates the Model Context Protocol (MCP) to enable AI-powered data analysis with interactive HTML dashboards.

Features

• MCP Integration: Connects to Incorta MCP Server for data querying and analysis
• Smart Prompt Caching: Implements Anthropic's best practices for 90% cost savings
• Session-Based Chat: Maintains conversation history across multiple interactions
• HTML Dashboards: Generates interactive visualizations using Chart.js and Tailwind CSS
• Dual-Task Architecture: Separates markdown analysis from HTML rendering

Installation

Upload Plugin to Incorta

./sdk/upload.sh incorta_mcp_plugin

Important: After uploading, restart Copilot with /restart in the chat window.

Configuration

1. CMC Plugin Configuration

Add to Cluster Configurations > IncortaCopilot > Plugins Configs:

{
  "IncortaMCPOperator": {
    "enabled": true,
    "clients": ["MATERIALIZED_VIEW", "BUSINESS_NOTEBOOK", "DASHBOARDS", "DASHBOARDS_V2"],
    "operator_name": "Incorta MCP Operator",
    "description": "Insert the MCP Capabilities to copilot using Incorta MCP",
    "operator_tag": "/incorta_claude",
    "operator_tag_description": "Use Incorta MCP to answer the user query",
    "operator_predefined_tasks": [
      {
        "id": 1,
        "operator": "IncortaMCPOperator",
        "operator_renderer": "MarkdownRenderer",
        "short_description": "Analyzing data..",
        "status": "TODO",
        "depends_on_output_of": [],
        "result": ""
      },
      {
        "id": 2,
        "operator": "IncortaMCPOperator",
        "operator_renderer": "HtmlRenderer",
        "short_description": "Processing results",
        "status": "TODO",
        "depends_on_output_of": [1],
        "result": ""
      },
      {
        "id": 3,
        "operator": "FinalResultText",
        "short_description": "Final Results and Insights",
        "status": "DONE",
        "depends_on_output_of": [1, 2],
        "result": "@task2 @task1"
      }
    ],
    "plugin_name": "incorta_mcp",
    "executor_args": {}
  }
}

Important: After updating configuration, restart Analytics service.

2. API Key Configuration

The plugin requires an Anthropic API key to function. You can configure it in one of two ways:

Option A: Environment Variable (Recommended)

Set the ANTHROPIC_API_KEY environment variable on the Analytics server:

export ANTHROPIC_API_KEY="sk-ant-api03-..."

Option B: CMC Configuration

Add the API key to the executor_args in CMC Plugins Configs:

{
  "IncortaMCPOperator": {
    "enabled": true,
    "executor_args": {
      "api_key": "sk-ant-api03-..."
    }
  }
}

Model Configuration: The plugin uses claude-sonnet-4-20250514 by default. This can be changed in the code if needed, but this has better limits.

3. MCP Server Configuration (Optional)

The plugin connects to the private endpoints version of the Incorta MCP server. By default, it uses a demo server, but you can configure your own:

{
  "IncortaMCPOperator": {
    "executor_args": {
      "mcp_server_url": "https://your-mcp-server.com/mcp/"
    }
  }
}

Default MCP Server: https://alone-recall-wait-era.trycloudflare.com/mcp/

The MCP server receives Incorta credentials from the user's session context automatically and in current setup we hardcode the SE environment internally.

4. Enable Plugin Upload

In CMC Advanced Configs:

{
  "general": {
    "allow_custom_plugins_upload": true,
    "allow_custom_plugins_execution": true
  }
}

5. Restart Copilot

• Go to Cloud Portal → Configuration
• Disable then re-enable Enable Copilot
• Or type /restart in the chat window

Usage

In the Copilot chat window:

/incorta_claude show me sales trends in a dashboard

The plugin will:

1. Task 1: Query data and generate analysis (with optional HTML dashboard code according to the context) → save response with extracting the HTML part separately and remove it from the main response
2. Task 2: Receive and render the HTML dashboard (if present), as we can set for it a specified renderer in frontend
3. Task 3: Display combined results (final)

Architecture

Resource Usage

Memory Storage Only: This plugin stores all data in memory (RAM), NOT on disk:

• Conversation history: Stored in Python global dictionary _conversation_history
• MCP client connections: Stored in memory
• Session data: In-memory only

Note: Anthropic's prompt caching is server-side (on Anthropic's infrastructure), not local disk caching.

File Structure

incorta_mcp_plugin/
├── __init__.py           # Package initialization
├── manifest.json         # Plugin metadata and version
├── plugin.py            # Main executor class
└── README.md            # This file

Key Components

1. IncortaMCPExecutor: Main execution class

- Manages MCP client connection - Handles LangGraph agent interactions - Implements smart caching logic

1. Session Management:

- Global conversation history per session - Prevents cache invalidation - Enables multi-turn conversations

1. HTML Extraction:

- Regex-based extraction from markdown - Document completeness validation - Truncation detection and logging

Requirements

Python packages (auto-installed):

• mcp
• anthropic
• langchain-core==0.3.75
• langchain==0.3.27
• langchain-anthropic==0.3.19
• langchain-mcp-adapters==0.1.9
• langgraph==0.2.60
• markdown

Authentication

Default credentials (can be overridden by context):

{
  "incorta_url": "https://se-prod-demo.cloud4.incorta.com/incorta",
  "tenant": "demo",
  "user": "admin",
  "password": "Incorta_1234%"
}

Changelog

Version 1.6.1

• Fixed prompt caching (system prompt only added once)
• Increased max_tokens to 8192 for complete dashboards
• Enhanced HTML validation with truncation detection
• Updated to production demo credentials
• Added context management

Troubleshooting

Plugin Not Appearing

Check logs for:

@log

Should show:

{
  "Available Plugins": [
    {
      "plugin_name": "incorta_mcp",
      "version": "1.6.1"
    }
  ]
}