Monti APM MCP Server

An MCP (Model Context Protocol) server that enables AI assistants to interact with Monti APM for Meteor application performance monitoring.

Features

• Query method execution traces with performance breakdowns
• Retrieve subscription/publication performance data
• Monitor HTTP request performance and errors
• Monitor system metrics (RAM, CPU, sessions, MongoDB pool)
• Track error rates and trends
• Analyze slow methods with actionable recommendations
• Identify performance bottlenecks across your app
• Get comprehensive health summaries with scores

Quick Start

1. Get Your Credentials

1. Log in to Monti APM Dashboard
2. Go to your app's Settings page
3. Copy your App ID and App Secret

2. Configure Claude Code

Add to your ~/.claude/mcp_servers.json:

{
  "mcpServers": {
    "montiapm": {
      "command": "npx",
      "args": ["@quave/montiapm-mcp"],
      "env": {
        "MONTI_APP_ID": "your-app-id-here",
        "MONTI_APP_SECRET": "your-app-secret-here"
      }
    }
  }
}

3. Restart Claude Code

After saving the config, restart Claude Code to load the MCP server.

4. Start Asking Questions

You can now ask Claude to analyze your Meteor app's performance:

"Give me a health summary of my application"
"What are the slowest methods in my app?"
"Show me error trends from the last hour"

Claude Code Subagent

Generate a specialized Claude Code subagent for Meteor performance analysis:

npx @quave/montiapm-mcp --generate-agent

This creates .claude/agents/meteor-performance.md with:

• Expert knowledge of Meteor performance optimization
• Pre-configured access to all Monti APM MCP tools
• Documentation-backed thresholds and recommendations
• Code examples for common optimization patterns

Options:

• --output <path> - Custom output path (default: .claude/agents/meteor-performance.md)
• --stdout - Print to stdout instead of file
• --force - Overwrite existing file

After running, Claude Code will automatically use this subagent when:

• Analyzing performance issues
• Investigating slow methods and publications
• Reviewing code changes for performance implications
• Getting optimization recommendations

How to Use

Diagnosing Slow Performance

Ask Claude to identify performance issues:

"My app feels slow. Can you analyze what's causing it?"
"Show me the top 10 slowest method calls from the last hour"
"Which methods are spending the most time in database operations?"
"Analyze performance bottlenecks in my app"

Investigating Specific Methods

Drill down into specific methods:

"Show me traces for the 'users.update' method"
"Get details for trace ID abc123"
"Why is my 'posts.list' method taking so long?"

Monitoring Publications

Analyze subscription performance:

"Show me slow publications"
"Which subscriptions are taking the longest to load?"
"Get traces for the 'userProfile' publication"

Monitoring HTTP Routes

Analyze HTTP request performance:

"Show me slow HTTP requests"
"Which API routes are taking the longest?"
"Get traces for the /api/users route"
"Are there any HTTP errors in my app?"

System Health

Check overall system health:

"What's my app's memory usage looking like?"
"Show me CPU usage over the last 2 hours"
"How many active sessions do I have?"
"Give me a complete health report"

Error Tracking

Investigate errors:

"Are there any error spikes recently?"
"Show me error trends for the last 24 hours"
"Is my error rate increasing or decreasing?"

Getting Recommendations

Get actionable advice:

"What should I optimize first in my app?"
"Give me recommendations for improving performance"
"Analyze slow methods and tell me how to fix them"

Available Tools

Trace Analysis

Tool	Description
get_method_traces	Retrieve method execution traces with time spent in DB, compute, HTTP, etc.
get_trace_detail	Get detailed events timeline for a specific trace
get_subscription_traces	Retrieve publication/subscription traces
get_http_traces	Retrieve HTTP request traces with performance metrics
get_error_traces	Retrieve error occurrence traces with filtering by type, status, and message
get_error_trace_detail	Get full details of a specific error including stack traces and client info

Metrics

Tool	Description
get_system_metrics	Get RAM, CPU, sessions, and MongoDB pool metrics
get_error_metrics	Get error count metrics and trends over time

Analysis

Tool	Description
analyze_slow_methods	Identify slow methods with optimization recommendations
analyze_performance_bottlenecks	Comprehensive bottleneck analysis across methods, pubs, and system
get_health_summary	Overall app health score (0-100) with insights

Example Conversations

Example 1: Quick Health Check

You: "How is my Meteor app doing?"

Claude: Uses get_health_summary tool and responds with:

• Health score: 85/100 (Good)
• Average response time: 145ms
• No errors in the last hour
• Memory usage is stable
• Insight: "Application is performing well with no immediate concerns."

Example 2: Investigating Slow Methods

You: "My users are complaining the app is slow. Help me find out why."

Claude: Uses analyze_slow_methods and analyze_performance_bottlenecks tools:

• Found 3 methods with response time > 500ms
• orders.search is the slowest at 2.3s average
• Main bottleneck: Database operations (78% of time)
• Recommendation: "Add indexes to the orders collection, particularly on the fields used in search queries"

Example 3: Debugging a Specific Issue

You: "The checkout flow is timing out. Check the 'cart.checkout' method."

Claude: Uses get_method_traces filtered by method name:

• Found 15 traces for cart.checkout in the last hour
• Average response time: 4.2s (very high!)
• Breakdown: DB: 3.1s, HTTP: 0.8s, Compute: 0.3s
• "The method is spending 74% of time in database operations. Let me get a detailed trace..."

Uses get_trace_detail:

• Timeline shows 12 sequential MongoDB queries
• Recommendation: "Batch these queries or use MongoDB aggregation pipeline"

Configuration Options

Environment Variables

Variable	Required	Description
MONTI_APP_ID	Yes	Your Monti APM application ID
MONTI_APP_SECRET	Yes	Your Monti APM application secret

Tool Parameters

Most tools accept these common parameters:

Parameter	Description	Default
startTime	Unix timestamp in milliseconds	1 hour ago
endTime	Unix timestamp in milliseconds	Now
limit	Maximum results to return	100

System Metrics Types

Available metrics for get_system_metrics:

• CPU_USAGE - CPU utilization percentage
• RAM_USAGE - Memory used by the app
• SESSIONS - Active DDP sessions
• NEW_SESSIONS - New sessions per minute
• MONGO_POOL_CHECKOUT_DELAY - MongoDB connection pool wait time

Resolution Options

For time-series data:

• RES_1MIN - 1-minute resolution (max 1000 minutes range)
• RES_30MIN - 30-minute resolution (max 14 days range)
• RES_3HOUR - 3-hour resolution (limited by plan retention)

Development

Prerequisites

• Node.js 18+
• npm

Setup

git clone https://github.com/quavedev/montiapm-mcp.git
cd montiapm-mcp
npm install

# Generate GraphQL types (required before building or running tests)
MONTI_APP_ID=xxx MONTI_APP_SECRET=xxx npm run codegen

Commands

# Run in development mode
npm run dev

# Build
npm run build

# Run unit tests
npm run test

# Run integration tests (requires credentials)
MONTI_APP_ID=xxx MONTI_APP_SECRET=xxx npm run test:integration

# Type check
npm run typecheck

# Generate GraphQL types (requires credentials)
MONTI_APP_ID=xxx MONTI_APP_SECRET=xxx npm run codegen

Project Structure

src/
├── auth/           # Authentication with Monti APM API
├── graphql/        # Apollo Client and GraphQL operations
│   ├── client.ts   # Apollo Client setup
│   ├── operations/ # GraphQL operation documents
│   └── generated/  # Auto-generated types (DO NOT EDIT)
├── tools/          # MCP tool implementations
├── utils/          # Shared utilities
├── server.ts       # MCP server setup
└── index.ts        # Entry point

API Rate Limits

• 5,000 requests/hour per app
• Max 1,000 records per query
• Resolution limits: 1min (1000 min range), 30min (14 days), 3hour (plan retention)

Troubleshooting

"Token expired or invalid"

• Verify your MONTI_APP_ID and MONTI_APP_SECRET are correct
• Check if credentials are properly passed to the MCP server

"Rate limit exceeded"

• API is limited to 5,000 requests/hour
• Reduce query frequency or increase time ranges

No data returned

• Check that your Meteor app is sending data to Monti APM
• Verify the time range includes periods with activity

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT

Related Links

• Monti APM Documentation
• Monti APM API Explorer
• Model Context Protocol