mcp-diet 0.1.0

Analyze MCP servers and see how much context window each tool consumes

mcp-diet

CLI tool that analyzes MCP servers and shows how much of your AI agent's context window each tool consumes, then grades efficiency.

Install

pip install mcp-diet
# or
uvx mcp-diet

Usage

# Analyze a running MCP server via stdio
mcp-diet analyze -- npx @modelcontextprotocol/server-filesystem /tmp

# Analyze via SSE
mcp-diet analyze --sse http://localhost:3000/sse

# Auto-detect from Claude config (.mcp.json or ~/.claude/settings.json)
mcp-diet auto

# Output formats
mcp-diet analyze --format json -- npx @modelcontextprotocol/server-filesystem /tmp
mcp-diet analyze --format markdown -- npx @modelcontextprotocol/server-filesystem /tmp

# Optimization suggestions
mcp-diet suggest -- npx @modelcontextprotocol/server-filesystem /tmp

What it shows

For each tool in the MCP server:

Column	Description
Tool name	The tool's registered name
Tokens	Estimated token count of the full tool definition
% of budget	Percentage of a configurable context budget (default: 200k)
Schema depth	Max nesting depth of the JSON schema
Desc length	Character count of the description
Grade	A-F rating based on token efficiency

Grading

Grade	Tokens per tool
A	< 200
B	200-500
C	500-1000
D	1000-2000
F	> 2000