MCP Servers

A collection of Model Context Protocol servers, templates, tools and more.

CLI tool to inspect MCP servers and analyze context token costs

Created 2/1/2026
Updated 2 days ago
Repository documentation and setup instructions

mcp-inspector

A CLI tool to inspect MCP (Model Context Protocol) servers and analyze their context token costs.

MCP servers add tools to your AI assistant, but each tool definition consumes context tokens. This tool helps you understand the "weight" of your MCP servers so you can optimize your setup and avoid context bloat.

Installation

npm install -g mcp-inspector

Or run directly with npx:

npx mcp-inspector

Usage

Inspect a Single Server

Inspect any MCP server by providing its command:

# Inspect the filesystem server
mcp-inspector inspect npx -y @modelcontextprotocol/server-filesystem /tmp

# Inspect a local server
mcp-inspector inspect node ./my-server.js

# Inspect with custom timeout
mcp-inspector inspect --timeout 60000 npx -y @modelcontextprotocol/server-postgres

# Get JSON output
mcp-inspector inspect --json npx -y @modelcontextprotocol/server-filesystem /tmp

Scan Your MCP Config

Scan all servers configured in your Claude Desktop or Cline config:

# Auto-detect config file
mcp-inspector scan

# Specify config file
mcp-inspector scan --config ~/.config/claude/claude_desktop_config.json

# JSON output
mcp-inspector scan --json

Help

mcp-inspector --help
mcp-inspector help-config

Output

The tool shows:

  • Server info: Name and version
  • Summary: Number of tools, resources, and prompts
  • Context cost: Estimated tokens for each component
  • Tools breakdown: Each tool with its token cost, sorted by weight
  • Recommendations: Suggestions when context cost is high

Example Output

Server Info
--------------------------------------------------
  Name:    filesystem
  Version: 1.0.0

Summary
--------------------------------------------------
  Tools:     5
  Resources: 0
  Prompts:   0

Context Cost (estimated tokens)
--------------------------------------------------
  Tools:     842
  Resources: 0
  Prompts:   0
  --------------------
  Total:     842

Tools
--------------------------------------------------
  read_file - 215 tokens
    Read the complete contents of a file from the file system
  write_file - 198 tokens
    Create a new file or overwrite an existing file with new content
  list_directory - 156 tokens
    Get a detailed listing of all files and directories in a specified path
  create_directory - 142 tokens
    Create a new directory or ensure a directory exists
  delete_file - 131 tokens
    Remove a file from the file system

Config Scan Output

Found 3 MCP server(s)

[filesystem]
Command: npx -y @modelcontextprotocol/server-filesystem /tmp
  Tools: 5, Resources: 0, Tokens: 842

[postgres]
Command: npx -y @modelcontextprotocol/server-postgres
  Tools: 3, Resources: 0, Tokens: 1.2k

[github]
Command: npx -y @modelcontextprotocol/server-github
  Tools: 12, Resources: 0, Tokens: 2.4k


Total Context Cost
--------------------------------------------------
  github                     2.4k (54%) ████████████
  postgres                   1.2k (27%) ██████
  filesystem                  842 (19%) ████
  ---------------------------------------------
  TOTAL                      4.4k

  Warning: High total context cost!
  Consider disabling unused servers or using ENABLE_TOOL_SEARCH=true

Why This Matters

Every MCP tool you add to Claude or other AI assistants consumes context tokens. This happens before your conversation even starts. If you have multiple servers with many tools, you might be burning thousands of tokens on tool definitions alone.

Common symptoms of context bloat:

  • Shorter effective conversations
  • Higher API costs
  • Model forgetting earlier context

Use mcp-inspector to:

  1. Identify which servers are the heaviest
  2. Decide which tools you actually need
  3. Optimize your MCP configuration

Tips for Reducing Context Cost

  1. Disable unused servers: If you don't use a server often, disable it
  2. Use ENABLE_TOOL_SEARCH=true: Claude can search tools instead of loading all definitions upfront
  3. Choose lightweight servers: Some servers expose fewer, more focused tools
  4. Build custom servers: Create servers with only the tools you need

Config File Locations

  • Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
  • Claude Desktop (Linux): ~/.config/claude/claude_desktop_config.json
  • Cline (VS Code): ~/.cline/mcp_settings.json

License

MIT

Quick Setup
Installation guide for this server

Install Package (if required)

npx @modelcontextprotocol/server-mcp-inspector

Cursor configuration (mcp.json)

{ "mcpServers": { "dabit3-mcp-inspector": { "command": "npx", "args": [ "dabit3-mcp-inspector" ] } } }