Bob Insights MCP Server

MCP server to analyze your local Bob AI conversation history. Identifies problematic conversations using regex pre-filtering and provides tools for detailed analysis.

Quick Start

Add to your Bob MCP settings (~/.bob/settings/mcp_settings.json):

{
  "mcpServers": {
    "bob-insights": {
      "command": "npx",
      "args": ["-y", "github:IBM/bob-insights"]
    }
  }
}

Restart Bob and try: "Can you list my recent conversations?"

Installation Options

Option 1: npx (Recommended)

No installation needed - uses latest from GitHub automatically.

Option 2: Global Install

npm install -g github:IBM/bob-insights

Then configure:

{
  "mcpServers": {
    "bob-insights": {
      "command": "bob-insights-mcp"
    }
  }
}

Option 3: From Source

git clone https://github.com/IBM/bob-insights.git
cd bob-insights
npm install && npm run build

Configure with local path:

{
  "mcpServers": {
    "bob-insights": {
      "command": "node",
      "args": ["/path/to/bob-insights/build/index.js"]
    }
  }
}

Available Tools

list_conversations

List recent conversations with problem scores.

Parameters:

• limit (optional): Number of conversations (default: 10, max: 100)
• days (optional): Only from last N days
• minScore (optional): Minimum problem score (default: 0)

Example: "Show me my last 20 conversations with problem scores above 30"

get_conversation

Get detailed information about a specific conversation.

Parameters:

• taskId (required): The conversation task ID

Example: "Show me details for conversation abc123xyz"

analyze_problems

Generate analysis prompt for problematic conversations.

Parameters:

• limit (optional): Number to check (default: 20)
• days (optional): Only from last N days
• minScore (optional): Minimum score to analyze (default: 30)
• question (optional): Custom analysis question

Example: "Analyze my problematic conversations from the last week"

Problem Scoring

Conversations are scored 0-100 based on:

Indicator	Points	Max
Tool failures	10 each	30
Error mentions	2 each	20
Retry attempts	3 each	15
Negative feedback	2 each	15
Confusion signals	2 each	10
Long conversation (>20 msgs)	5	5
High tool usage (>15 tools)	5	5

Severity:

• 70-100: High
• 40-69: Medium
• 30-39: Low
• 0-29: Smooth

How It Works

1. Automatically finds Bob conversation history in local storage
2. Uses regex to detect problem indicators (errors, retries, negative feedback)
3. Calculates problem scores (0-100)
4. Provides tools for Bob to analyze conversations
5. Reduces LLM analysis costs by 80-90% through smart filtering

Storage Locations

• Windows: %APPDATA%\Bob-IDE\User\globalStorage\ibm.bob-code\tasks
• macOS: ~/Library/Application Support/Bob-IDE/User/globalStorage/ibm.bob-code/tasks

Development

npm install
npm run build      # Compile TypeScript
npm run watch      # Watch mode

License

MIT