Table of Contents
- Overview
- Difference Between the MCP Inspector and MCP Tools
- Installation
- Getting Started
- Features
- Server Aliases
- LLM Apps Config Management
- Server Modes
- Examples
- Contributing
- Roadmap
- License
Overview
MCP Tools provides a versatile CLI for working with Model Context Protocol (MCP) servers. It enables you to:
- Discover and call tools provided by MCP servers
- Access and utilize resources exposed by MCP servers
- Create mock servers for testing client applications
- Proxy MCP requests to shell scripts for easy extensibility
- Create interactive shells for exploring and using MCP servers
- Scaffold new MCP projects with TypeScript support
- Format output in various styles (JSON, pretty-printed, table)
- Guard and restrict access to specific tools and resources
- Support all transport methods (HTTP, stdio)
Installation
Using Homebrew (for macOS)
brew tap f/mcptools
brew install mcp
❕ The binary is installed as
mcpbut can also be accessed asmcptto avoid conflicts with other tools that might use themcpcommand name.
From Source (for Windows and GNU/Linux)
go install github.com/f/mcptools/cmd/mcptools@latest
❕ The binary will be installed as
mcptoolswhen but can be aliased tomcptfor convenience.Windows 11 Running Example
Getting Started
The simplest way to start using MCP Tools is to connect to an MCP server and list available tools:
# List all available tools from a filesystem server
mcp tools npx -y @modelcontextprotocol/server-filesystem ~
# Call a specific tool
mcp call read_file --params '{"path":"README.md"}' npx -y @modelcontextprotocol/server-filesystem ~
# Open an interactive shell
mcp shell npx -y @modelcontextprotocol/server-filesystem ~
Features
MCP Tools supports a wide range of features for interacting with MCP servers:
Usage:
mcp [command]
Available Commands:
version Print the version information
tools List available tools on the MCP server
resources List available resources on the MCP server
prompts List available prompts on the MCP server
call Call a tool, resource, or prompt on the MCP server
get-prompt Get a prompt on the MCP server
read-resource Read a resource on the MCP server
shell Start an interactive shell for MCP commands
web Start a web interface for MCP commands
mock Create a mock MCP server with tools, prompts, and resources
proxy Proxy MCP tool requests to shell scripts
alias Manage MCP server aliases
configs Manage MCP server configurations
new Create a new MCP project component
help Help about any command
completion Generate the autocompletion script for the specified shell
Flags:
-f, --format string Output format (table, json, pretty) (default "table")
-h, --help help for mcp
-p, --params string JSON string of parameters to pass to the tool (for call command) (default "{}")
Use "mcp [command] --help" for more information about a command.
Transport Options
MCP Tools supports multiple transport methods for communicating with MCP servers:
Stdio Transport
Uses stdin/stdout to communicate with an MCP server via JSON-RPC 2.0. This is useful for command-line tools that implement the MCP protocol.
mcp tools npx -y @modelcontextprotocol/server-filesystem ~
HTTP SSE Transport
Uses HTTP and Server-Sent Events (SSE) to communicate with an MCP server via JSON-RPC 2.0. This is useful for connecting to remote servers that implement the legacy MCP protocol. Transport is automatically detected when the URL ends with /sse.
mcp tools http://localhost:3001/sse
# Example: Use the everything sample server
# docker run -p 3001:3001 --rm -it tzolov/mcp-everything-server:v1
Note: HTTP SSE currently supports only MCP protocol version 2024-11-05.
Streamable HTTP Transport (Recommended)
Uses streamable HTTP to communicate with an MCP server via JSON-RPC 2.0. This is the modern, recommended approach for connecting to remote servers that implement the MCP protocol. It supports both streaming responses and simple request/response patterns. This is the default transport for HTTP/HTTPS URLs.
# Default transport for HTTP URLs
mcp tools http://localhost:3000
# Streamable HTTP transport (auto-detected from URL)
mcp tools http://localhost:3000
# Examples with remote servers
mcp tools https://api.example.com/mcp
mcp tools https://ne.tools
Benefits of Streamable HTTP:
- Session Management: Supports stateful connections with session IDs
- Resumability: Can reconnect and resume interrupted sessions (when supported by server)
- Flexible Responses: Supports both streaming and direct JSON responses
- Modern Protocol: Uses the latest MCP transport specification
Output Formats
MCP Tools supports three output formats to accommodate different needs:
Table Format (Default)
mcp tools npx -y @modelcontextprotocol/server-filesystem ~
The default format now displays tools in a colorized man-page style:
read_file(path:str)
Read the complete contents of a file from the file system.
read_multiple_files(paths:str[])
Read the contents of multiple files simultaneously.
list_dir(path:str)
Lists the contents of a directory.
write_file(path:str, content:str)
Writes content to a file.
grep_search(pattern:str, [excludePatterns:str[]])
Search files with pattern.
edit_file(edits:{newText:str,oldText:str}[], path:str)
Edit a file with multiple text replacements
Key features of the format:
- Function names are displayed in bold cyan
- Required parameters are shown in green (e.g.,
path:str) - Optional parameters are shown in yellow brackets (e.g.,
[limit:int]) - Array types are indicated with
[]suffix (e.g.,str[]) - Object types show their properties in curly braces (e.g.,
{prop1:type1,prop2:type2}) - Nested objects are displayed recursively (e.g.,
{notifications:{enabled:bool,sound:bool}}) - Type names are shortened for readability (e.g.,
strinstead ofstring,intinstead ofinteger) - Descriptions are indented and displayed in gray
- Parameter order is consistent, with required parameters listed first
JSON Format (Compact)
mcp tools --format json npx -y @modelcontextprotocol/server-filesystem ~
Pretty JSON Format (Indented)
mcp tools --format pretty npx -y @modelcontextprotocol/server-filesystem ~
Commands
MCP Tools includes several core commands for interacting with MCP servers:
List Available Tools
mcp tools npx -y @modelcontextprotocol/server-filesystem ~
List Available Resources
mcp resources npx -y @modelcontextprotocol/server-filesystem ~
List Available Prompts
mcp prompts npx -y @modelcontextprotocol/server-filesystem ~
Call a Tool
mcp call read_file --params '{"path":"/path/to/file"}' npx -y @modelcontextprotocol/server-filesystem ~
Call a Resource
mcp call resource:test://static/resource/1 npx -y @modelcontextprotocol/server-everything -f json | jq ".contents[0].text"
or
mcp read-resource test://static/resource/1 npx -y @modelcontextprotocol/server-everything -f json | jq ".contents[0].text"
Call a Prompt
mcp get-prompt simple_prompt npx -y @modelcontextprotocol/server-everything -f json | jq ".messages[0].content.text"
Viewing Server Logs
When using client commands that make calls to the server, you can add the --server-logs flag to see the server logs related to your request:
# View server logs when listing tools
mcp tools --server-logs npx -y @modelcontextprotocol/server-filesystem ~
Output:
[>] Secure MCP Filesystem Server running on stdio
[>] Allowed directories: [ '/Users/fka/' ]
read_file(path:str)
Read the complete contents of a file from the file system.
read_multiple_files(paths:str[])
Read the contents of multiple files simultaneously.
... and the other tools available on this server
This can be helpful for debugging or understanding what's happening on the server side when executing these commands.
Interactive Shell
The interactive shell mode allows you to run multiple MCP commands in a single session:
mcp shell npx -y @modelcontextprotocol/server-filesystem ~
This opens an interactive shell with the following capabilities:
mcp tools shell
connected to: npx -y @modelcontextprotocol/server-filesystem /Users/fka
mcp > Type '/h' for help or '/q' to quit
mcp > tools
read_file(path:str, [limit:int], [offset:int])
Reads a file from the filesystem
list_dir(path:str)
Lists directory contents
grep_search(pattern:str, [excludePatterns:str[]])
Search files with pattern
edit_file(edits:{newText:str,oldText:str}[], path:str)
Edit a file with multiple text replacements
# Direct tool calling is supported
mcp > read_file {"path":"README.md"}
...content of README.md...
# Calling a tool with complex object parameters
mcp > edit_file {"path":"main.go","edits":[{"oldText":"foo","newText":"bar"}]}
...result of edit operation...
# Get help
mcp > /h
MCP Shell Commands:
tools List available tools
resources List available resources
prompts List available prompts
call <entity> [--params '{...}'] Call a tool, resource, or prompt
format [json|pretty|table] Get or set output format
Special Commands:
/h, /help Show this help
/q, /quit, exit Exit the shell
Web Interface
MCP Tools provides a web interface for interacting with MCP servers through a browser-based UI:
# Start a web interface for a filesystem server on default port (41999)
mcp web npx -y @modelcontextprotocol/server-filesystem ~
# Use a custom port
mcp web --port 8080 docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server
# Use SSE
mcp web https://ne.tools
The web interface includes:
- A sidebar listing all available tools, resources, and prompts
- Form-based and JSON-based parameter editing
- Formatted and raw JSON response views
- Interactive parameter forms automatically generated from tool schemas
- Support for complex parameter types (arrays, objects, nested structures)
- Direct API access for tool calling
Once started, you can access the interface by opening http://localhost:41999 (or your custom port) in a browser.
Project Scaffolding
MCP Tools provides a scaffolding feature to quickly create new MCP servers with TypeScript:
mkdir my-mcp-server
cd my-mcp-server
# Create a project with specific components
mcp new tool:calculate resource:file prompt:greet
# Create a project with a specific SDK (currently only TypeScript/ts supported)
mcp new tool:calculate --sdk=ts
# Create a project with a specific transport type
mcp new tool:calculate --transport=stdio
mcp new tool:calculate --transport=sse
The scaffolding creates a complete project structure with:
- Server setup with chosen transport (stdio or SSE)
- TypeScript configuration with modern ES modules
- Component implementations with proper MCP interfaces
- Automatic wiring of imports and initialization
After scaffolding, you can build and run your MCP server:
# Install dependencies
npm install
# Build the TypeScript code
npm run build
# Test the server with MCP Tools
mcp tools node build/index.js
Project templates are stored in either:
- Local
./templates/directory - User's home directory:
~/.mcpt/templates/ - Homebrew installation path (
/opt/homebrew/Cellar/mcp/v#.#.#/templates)
When installing via Homebrew, templates are automatically installed to your home directory. But if you use source install, you need to run make install-templates.
Server Aliases
MCP Tools allows you to save and reuse server commands with friendly aliases:
# Add a new server alias
mcp alias add myfs npx -y @modelcontextprotocol/server-filesystem ~/
# List all registered server aliases
mcp alias list
# Remove a server alias
mcp alias remove myfs
# Use an alias with any MCP command
mcp tools myfs
mcp call read_file --params '{"path":"README.md"}' myfs
Server aliases are stored in $HOME/.mcpt/aliases.json and provide a convenient way to work with commonly used MCP servers without typing long commands repeatedly.
LLM Apps Config Management
MCP Tools provides a powerful configuration management system that helps you work with MCP server configurations across multiple applications:
🚧 This works only on macOS for now.
# Scan for MCP server configurations across all supported applications
mcp configs scan
# List all configurations (alias for configs view --all)
mcp configs ls
# View specific configuration by alias
mcp configs view vscode
# Add or update a server in a configuration
mcp configs set vscode my-server npm run mcp-server
mcp configs set cursor my-api https://api.example.com/mcp --headers "Authorization=Bearer token"
# Add to multiple configurations at once
mcp configs set vscode,cursor,claude-desktop my-server npm run mcp-server
# Remove a server from a configuration
mcp configs remove vscode my-server
# Create an alias for a custom config file
mcp configs alias myapp ~/myapp/config.json
# Synchronize and merge configurations from multiple sources
mcp configs sync vscode cursor --output vscode --default interactive
# Convert a command line to MCP server JSON configuration format
mcp configs as-json mcp proxy start
# Output: {"command":"mcp","args":["proxy","start"]}
# Convert a URL to MCP server JSON configuration format
mcp configs as-json https://api.example.com/mcp --headers "Authorization=Bearer token"
# Output: {"url":"https://api.example.com/mcp","headers":{"Authorization":"Bearer token"}}
Configurations are managed through a central registry in $HOME/.mcpt/configs.json with predefined aliases for:
- VS Code and VS Code Insiders
- Windsurf
- Cursor
- Claude Desktop and Claude Code
The system automatically displays server configurations in a colorized format grouped by source, showing command-line or URL information, headers, and environment variables.
mcp configs scan command looks for MCP server configurations in:
- Visual Studio Code
- Visual Studio Code Insiders
- Windsurf
- Cursor
- Claude Desktop
Example Output:
VS Code Insiders
GitHub (stdio):
docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server
Claude Desktop
Proxy (stdio):
mcp proxy start
My Files (stdio):
npx -y @modelcontextprotocol/server-filesystem ~/
Bonus
Add official GitHub MCP Server to Windsurf, Cursor and VS Code at once:
mcp configs set windsurf,cursor,vscode GitHub \
--env "GITHUB_PERSONAL_ACCESS_TOKEN=github_pat_xxx" \
docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server
Server Modes
MCP Tools can operate as both a client and a server, with two server modes available:
Mock Server Mode
The mock server mode creates a simulated MCP server for testing clients without implementing a full server:
# Create a mock server with a simple tool
mcp mock tool hello_world "A simple greeting tool"
# Create a mock server with multiple entity types
mcp mock tool hello_world "A greeting tool" \
prompt welcome "A welcome prompt" "Hello {{name}}, welcome to {{location}}!" \
resource docs://readme "Documentation" "Mock MCP Server\nThis is a mock server"
Features of the mock server:
- Full initialization handshake
- Tool listing with standardized schema
- Tool calling with simple responses
- Resource listing and reading
- Prompt listing and retrieval with argument substitution
- Detailed request/response logging to
~/.mcpt/logs/mock.log
Using Prompt Templates
For prompts, any text in {{double_braces}} is automatically detected as an argument:
# Create a prompt with name and location arguments
mcp mock prompt greeting "Greeting template" "Hello {{name}}! Welcome to {{location}}."
When a client requests the prompt, it can provide values for these arguments which will be substituted in the response.
Proxy Mode
The proxy mode allows you to register shell scripts or inline commands as MCP tools, making it easy to extend MCP functionality without writing code:
# Register a shell script as an MCP tool
mcp proxy tool add_operation "Adds a and b" "a:int,b:int" ./examples/add.sh
# Register an inline command as an MCP tool
mcp proxy tool add_operation "Adds a and b" "a:int,b:int" -e 'echo "total is $a + $b = $(($a+$b))"'
# Register an inline command as an MCP tool with optional parameter
mcpt proxy tool add_operation "Adds a and b with optional result msg" "a:int,b:int,[msg:string]" -e 'echo "$msg$a + $b = $(($a+$b))"'
# Unregister a tool
mcp proxy tool --unregister add_operation
# Start the proxy server
mcp proxy start
Running mcp tools localhost:3000 with the proxy server will show the registered tools with their parameters:
add_operation(a:int, b:int)
Adds a and b
count_files(dir:str, [include:str[]])
Counts files in a directory with optional filters
This new format clearly shows what parameters each tool accepts, making it easier to understand how to use them. Arrays are denoted with [] suffix (e.g., str[]), and type names are shortened for better readability.
How It Works
- Register a shell script or inline command with a tool name, description, and parameter specification
- Start the proxy server, which implements the MCP protocol
- When a tool is called, parameters are passed as environment variables to the script/command
- The script/command's output is returned as the tool response
- If the script's output is a base64-encoded PNG image (prefixed with
data:image/png;base64,), it is returned as an ImageContent object.
Example Scripts and Commands
Adding Numbers (add.sh):
#!/bin/bash
# Get values from environment variables
if [ -z "$a" ] || [ -z "$b" ]; then
echo "Error: Missing required parameters 'a' or 'b'"
exit 1
fi
# Perform the addition
result=$(($a + $b))
echo "The sum of $a and $b is $result"
Generating a QR Code
This example requires a tool like qrencode to be installed.
# Register a tool to generate a QR code
mcp proxy tool qrcode "Generates a QR code" "text:string" \
-e 'echo -e "data:image/png;base64,$(qrencode -t png -o - "$text" | base64 -w 0)"'
Inline Command Example:
# Simple addition
mcp proxy tool add_op "Adds given numbers" "a:int,b:int" -e 'echo "total is $a + $b = $(($a+$b))"'
# Customized greeting
mcp proxy tool greet "Greets a user" "name:string,greeting:string,formal:bool" -e '
if [ "$formal" = "true" ]; then
title="Mr./Ms."
echo "${greeting:-Hello}, ${title} ${name}. How may I assist you today?"
else
echo "${greeting:-Hello}, ${name}! Nice to meet you!"
fi
'
# File operations
mcp proxy tool count_lines "Counts lines in a file" "file:string" -e "wc -l < \"$file\""
Configuration and Logging
- Tools are registered in
~/.mcpt/proxy_config.json - The proxy server logs all requests and responses to
~/.mcpt/logs/proxy.log - Use
--unregisterto remove a tool from the configuration
Guard Mode
The guard mode allows you to restrict access to specific tools, prompts, and resources based on pattern matching. This is useful for security purposes when:
- Restricting potentially dangerous operations (file writes, deletions, etc.)
- Limiting the capabilities of AI assistants or applications
- Providing read-only access to sensitive systems
- Creating sandboxed environments for testing or demonstrations
Note: Guard mode currently only works with STDIO transport (command execution) and not HTTP transport.
# Allow only file reading operations, deny file modifications
mcp guard --allow 'tools:read_* --deny tools:write_*,create_*,delete_*' npx -y @modelcontextprotocol/server-filesystem ~
# Permit only a single specific tool
mcp guard --allow 'tools:search_files' npx -y @modelcontextprotocol/server-filesystem ~
# Restrict by both tool type and prompt type
mcp guard --allow 'tools:read_*,prompts:system_*' --deny tools:execute_* npx -y @modelcontextprotocol/server-filesystem ~
# Using with aliases
mcp guard --allow 'tools:read_*' fs # Where 'fs' is an alias for a filesystem server
How It Works
The guard command works by:
- Creating a proxy that sits between the client and the MCP server
- Intercepting and filtering all requests to
tools/list,prompts/list, andresources/list - Preventing calls to tools, prompts, or resources that don't match the allowed patterns
- Blocking requests for filtered resources, tools and prompts
- Passing through all other requests and responses unchanged
Pattern Matching
Patterns use simple glob syntax with * as a wildcard:
tools:read_*- Matches all tools starting with "read_"tools:*file*- Matches any tool with "file" in the nameprompts:system_*- Matches all prompts starting with "system_"
For each entity type, you can specify:
--allow 'pattern1,pattern2,...'- Only allow entities matching these patterns--deny 'pattern1,pattern2,...'- Remove entities matching these patterns
If no allow patterns are specified, all entities are allowed by default (except those matching deny patterns).
Application Integration
You can use the guard command to secure MCP configurations in applications. For example, to restrict a file system server to only allow read operations, change:
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/Users/fka/Desktop"
]
}
To:
"filesystem": {
"command": "mcp",
"args": [
"guard", "--deny", "tools:write_*,create_*,move_*,delete_*",
"npx", "-y", "@modelcontextprotocol/server-filesystem",
"/Users/fka/Desktop"
]
}
This provides a read-only view of the filesystem by preventing any modification operations.
You can also use aliases with the guard command in configurations:
"filesystem": {
"command": "mcp",
"args": [
"guard", "--allow", "tools:read_*,list_*,search_*",
"fs" // Where 'fs' is an alias for the filesystem server
]
}
This makes your configurations even more concise and easier to maintain.
Logging
- Guard operations are logged to
~/.mcpt/logs/guard.log - The log includes all requests, responses, and filtering decisions
- Use
tail -f ~/.mcpt/logs/guard.logto monitor activity in real-time
Examples
Basic Usage
List tools from a filesystem server:
mcp tools npx -y @modelcontextprotocol/server-filesystem ~
Call a tool with pretty JSON output:
mcp call read_file --params '{"path":"README.md"}' --format pretty npx -y @modelcontextprotocol/server-filesystem ~
Use the guard mode to filter available tools:
# Only allow file search functionality
mcp guard --allow tools:search_files npx -y @modelcontextprotocol/server-filesystem ~
# Create a read-only environment
mcp guard --deny tools:write_*,delete_*,create_*,move_* npx -y @modelcontextprotocol/server-filesystem ~
Streamable HTTP Usage
Create and run a local streamable HTTP server:
# Create a new MCP server with SSE transport (streamable HTTP not available in scaffolding)
mkdir my-sse-server && cd my-sse-server
mcp new tool:example_tool --transport=sse
# Install dependencies and build
npm install && npm run build
# Start the server (will run SSE transport)
npm start
In a separate terminal, connect to your local server:
# Connect to local SSE server (adjust URL based on your server's SSE endpoint)
mcp tools http://localhost:3000/sse
# Call a tool on the local server
mcp call example_tool --params '{"input": "test"}' http://localhost:3000/sse
# Use with different output formats
mcp tools --format pretty http://localhost:3000/sse
Connect to remote streamable HTTP servers:
# Connect to a remote MCP server
mcp tools https://api.example.com/mcp
# Use SSE transport for legacy servers (auto-detected from /sse path)
mcp tools http://legacy-server.com/sse
# Example with authentication headers (when supported)
mcp tools https://authenticated-mcp-server.com
Script Integration
Using the proxy mode with a simple shell script:
# 1. Create a simple shell script for addition
cat > add.sh << 'EOF'
#!/bin/bash
# Get values from environment variables
if [ -z "$a" ] || [ -z "$b" ]; then
echo "Error: Missing required parameters 'a' or 'b'"
exit 1
fi
result=$(($a + $b))
echo "The sum of $a and $b is $result"
EOF
# 2. Make it executable
chmod +x add.sh
# 3. Register it as an MCP tool
mcp proxy tool add_numbers "Adds two numbers" "a:int,b:int" ./add.sh
# 4. In one terminal, start the proxy server
mcp proxy start
# 5. In another terminal, you can call it as an MCP tool
mcp call add_numbers --params '{"a":5,"b":3}' --format pretty
Debugging
Tailing the logs to debug your proxy or mock server:
# For the mock server logs
tail -f ~/.mcpt/logs/mock.log
# For the proxy server logs
tail -f ~/.mcpt/logs/proxy.log
# To watch all logs in real-time (on macOS/Linux)
find ~/.mcpt/logs -name "*.log" -exec tail -f {} \;
Contributing
We welcome contributions! Please see our Contributing Guidelines for details on how to submit pull requests, report issues, and contribute to the project.
Roadmap
The following features are planned for future releases:
- Authentication: Support for secure authentication mechanisms
License
This project is licensed under the MIT License.
Thanks
Thanks to Fatih Taskiran for the logo design.