mcp tokens
Estimate MCP token usage for server capabilities by connecting to servers and analyzing their tools, resources, and prompts.
For a complete overview of server management, see the Server Management Guide.
Synopsis
npx -y @1mcp/agent mcp tokens [options]
Options
--model, -m <model>
- Model to use for token estimation (default:
gpt-4o
) - Supports any valid tiktoken model including
gpt-3.5-turbo
,gpt-4
,o1
, etc.
- Model to use for token estimation (default:
--preset, -p <name>
- Use preset filter instead of manual tag expression
- Cannot be used together with
--tag-filter
--tag-filter, -f <expression>
- Filter servers by advanced tag expression (and/or/not logic)
- Examples:
network
,web+api
,(filesystem,network)-test
- Cannot be used together with
--preset
--format <format>
- Output format:
table
,json
,summary
(default:table
)
- Output format:
--verbose, -v
- Show server logs and connection details
- By default, server logs are suppressed for clean output
--config, -c <path>
- Path to the config file
Description
The mcp tokens
command connects to your configured MCP servers and analyzes their capabilities to estimate token usage. This is useful for:
- Cost Planning: Estimate token costs when using different language models
- Performance Analysis: Understand which servers contribute most to token usage
- Optimization: Identify opportunities to reduce token consumption
- Debugging: Analyze server connectivity and capability loading
The command uses tiktoken for accurate tokenization based on the specified model, with automatic fallback handling for invalid models.
Examples
# Estimate tokens for all MCP servers using default model (gpt-4o)
npx -y @1mcp/agent mcp tokens
# Use a different model for token estimation
npx -y @1mcp/agent mcp tokens --model gpt-3.5-turbo
# Use a preset for server filtering
npx -y @1mcp/agent mcp tokens --preset development
# Filter servers by tags and show summary format
npx -y @1mcp/agent mcp tokens --tag-filter="network or filesystem" --format=summary
# Export detailed analysis as JSON for programmatic use
npx -y @1mcp/agent mcp tokens --format=json > token-analysis.json
# Complex tag filtering with specific model
npx -y @1mcp/agent mcp tokens --model=o1 --tag-filter="(web+api)-test" --format=table
# Show server logs and connection details for debugging
npx -y @1mcp/agent mcp tokens --verbose
# Use custom config file
npx -y @1mcp/agent mcp tokens --config ~/my-mcp-config.json
Output Formats
Table Format (Default)
Shows a beautifully formatted, colorful hierarchical breakdown by capability type (Tools, Resources, Prompts) with server groupings and token estimates for each item. The output uses:
- Colored sections with visual icons (🔧 Tools, 📁 Resources, 💬 Prompts)
- Boxed summary with overall statistics
- Clean layout with server logs suppressed by default
- Connection status clearly indicated for each server
JSON Format
Provides structured data suitable for programmatic analysis:
{
"summary": {
"totalServers": 5,
"connectedServers": 4,
"totalTools": 25,
"overallTokens": 8450
},
"servers": [
{
"serverName": "filesystem",
"connected": true,
"breakdown": {
"tools": [...],
"totalTokens": 1200
}
}
]
}
Summary Format
Concise overview with key metrics and top servers by token usage, presented in colorful boxed sections:
- Usage summary with server counts and capability totals
- Top servers ranking by token consumption
- Connection issues clearly highlighted if any servers are down
Supported Models
The --model
option accepts any valid tiktoken model:
- GPT-4 Family:
gpt-4o
,gpt-4
,gpt-4-turbo
- GPT-3.5 Family:
gpt-3.5-turbo
,gpt-3.5-turbo-16k
- O1 Family:
o1
,o1-mini
,o1-preview
- Legacy Models:
text-davinci-003
,code-davinci-002
If an invalid model is specified, the command automatically falls back to gpt-4o
with a warning.
Error Handling
- Server Connection Failures: Displayed in output with error details
- Invalid Model Names: Automatic fallback to
gpt-4o
- Configuration Issues: Clear error messages with suggested fixes
- Permission Errors: Helpful guidance for authentication issues
See Also
- Server Management Guide
- mcp status - Check server connectivity
- mcp list - List configured servers