MCP CLI Tool

mcp-cli Tool

Skill by ara.so — MCP Skills collection.

Overview

mcp-cli is a lightweight, Bun-based CLI for interacting with MCP (Model Context Protocol) servers. It provides:

• On-demand schema loading - Only fetch tool schemas when needed, saving AI context tokens
• Shell composability - JSON output for piping with jq, chaining, and scripting
• Connection pooling - Lazy-spawn daemon keeps connections warm (60s idle timeout)
• Universal support - Works with both stdio and HTTP MCP servers
• Tool filtering - Allow/disable specific tools per server via config

Use cases:

• AI agents accessing MCP tools without loading full schemas into context
• Shell scripts automating MCP server interactions
• Discovering and exploring available MCP capabilities

Installation

Quick Install (Recommended)

curl -fsSL https://raw.githubusercontent.com/philschmid/mcp-cli/main/install.sh | bash

Manual Install (Requires Bun)

bun install -g https://github.com/philschmid/mcp-cli

Verify Installation

mcp-cli --version

Configuration

Config File Location

Create mcp_servers.json in one of these locations (searched in order):

1. Path from MCP_CONFIG_PATH environment variable
2. Path from -c/--config CLI argument
3. ./mcp_servers.json (current directory)
4. ~/.mcp_servers.json
5. ~/.config/mcp/mcp_servers.json

Basic Config Format

Compatible with Claude Desktop, Gemini, and VS Code:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ],
      "env": {
        "GITHUB_TOKEN": "${GITHUB_TOKEN}"
      }
    },
    "deepwiki": {
      "url": "https://mcp.deepwiki.com/mcp"
    }
  }
}

Advanced Config Options

{
  "mcpServers": {
    "local-server": {
      "command": "node",
      "args": ["./server.js"],
      "env": {
        "API_KEY": "${API_KEY}"
      },
      "cwd": "/path/to/directory",
      "allowedTools": ["read_*", "list_*"],
      "disabledTools": ["delete_*"]
    },
    "remote-server": {
      "url": "https://mcp.example.com",
      "headers": {
        "Authorization": "Bearer ${TOKEN}"
      }
    }
  }
}

Environment Variable Substitution

Use ${VAR_NAME} syntax anywhere in config:

{
  "mcpServers": {
    "api-server": {
      "command": "node",
      "args": ["server.js"],
      "env": {
        "DATABASE_URL": "${DATABASE_URL}",
        "API_KEY": "${API_KEY}"
      }
    }
  }
}

Control behavior:

• MCP_STRICT_ENV=true (default): Error on missing variables
• MCP_STRICT_ENV=false: Use empty values with warning

Tool Filtering

Restrict available tools using glob patterns:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."],
      "allowedTools": ["read_file", "list_directory"],
      "disabledTools": ["delete_file"]
    }
  }
}

Rules:

• allowedTools: Whitelist (supports *, ? glob patterns)
• disabledTools: Blacklist (takes precedence over allowedTools)
• Applies to all operations (info, grep, call)

Examples:

// Only read operations
"allowedTools": ["read_*", "list_*", "search_*"]

// Disable destructive operations
"disabledTools": ["delete_*", "write_*", "create_*"]

// Combine filters
"allowedTools": ["*file*"],
"disabledTools": ["delete_file"]

Core Commands

List All Servers and Tools

# Basic listing
mcp-cli

# With descriptions
mcp-cli -d

Output:

github
  • search_repositories
  • get_file_contents
  • create_or_update_file
filesystem
  • read_file
  • write_file
  • list_directory

Search Tools by Pattern

# Find file-related tools
mcp-cli grep "*file*"

# Search with descriptions
mcp-cli grep "*search*" -d

Output:

github/get_file_contents
github/create_or_update_file
filesystem/read_file
filesystem/write_file

View Server Details

mcp-cli info <server>

Example:

mcp-cli info github

Output:

Server: github
Transport: stdio
Command: npx -y @modelcontextprotocol/server-github

Tools (12):
  search_repositories
    Search for GitHub repositories
    Parameters:
      • query (string, required) - Search query
      • page (number, optional) - Page number
  ...

View Tool Schema

Both formats work:

mcp-cli info <server> <tool>
mcp-cli info <server>/<tool>

Example:

mcp-cli info github search_repositories

Output:

Tool: search_repositories
Server: github

Description:
  Search for GitHub repositories

Input Schema:
  {
    "type": "object",
    "properties": {
      "query": { "type": "string", "description": "Search query" },
      "page": { "type": "number" }
    },
    "required": ["query"]
  }

Call a Tool

# Inline JSON
mcp-cli call <server> <tool> '{"key": "value"}'

# From stdin (auto-detected, no '-' needed)
echo '{"path": "./file"}' | mcp-cli call <server> <tool>

# Heredoc for complex JSON
mcp-cli call <server> <tool> <<EOF
{"content": "Text with 'quotes' and \"escapes\""}
EOF

Example:

mcp-cli call github search_repositories '{"query": "mcp server", "per_page": 5}'

Output (JSON):

{
  "content": [
    {
      "type": "text",
      "text": "{\"items\": [{\"name\": \"mcp-cli\", \"url\": \"...\"}]}"
    }
  ]
}

Practical Usage Patterns

1. Discover → Inspect → Execute Workflow

# Step 1: List available servers
mcp-cli

# Step 2: View server tools
mcp-cli info filesystem

# Step 3: Get tool schema
mcp-cli info filesystem read_file

# Step 4: Call the tool
mcp-cli call filesystem read_file '{"path": "./README.md"}'

2. Pipe JSON with jq

# Extract specific field
mcp-cli call github search_repositories '{"query": "mcp"}' | jq '.content[0].text'

# Parse nested JSON
mcp-cli call github search_repositories '{"query": "mcp"}' \
  | jq -r '.content[0].text | fromjson | .items[].html_url'

# Filter results
mcp-cli call filesystem list_directory '{"path": "."}' \
  | jq -r '.content[0].text | split("\n")[] | select(endswith(".md"))'

3. Chain Multiple MCP Calls

# Search and read first result
mcp-cli call filesystem search_files '{"path": "src/", "pattern": "*.ts"}' \
  | jq -r '.content[0].text | split("\n")[0]' \
  | xargs -I {} mcp-cli call filesystem read_file '{"path": "{}"}'

# Process all matching files
mcp-cli call filesystem search_files '{"path": ".", "pattern": "*.md"}' \
  | jq -r '.content[0].text | split("\n")[]' \
  | while read file; do
      echo "=== $file ==="
      mcp-cli call filesystem read_file "{\"path\": \"$file\"}" | jq -r '.content[0].text'
    done

4. Error Handling in Scripts

# Conditional execution
mcp-cli call filesystem list_directory '{"path": "."}' \
  | jq -e '.content[0].text | contains("README.md")' \
  && mcp-cli call filesystem read_file '{"path": "./README.md"}'

# Fallback on error
if result=$(mcp-cli call filesystem read_file '{"path": "./config.json"}' 2>/dev/null); then
  echo "$result" | jq '.content[0].text | fromjson'
else
  echo "File not found, using defaults"
fi

5. Save Output to File

mcp-cli call github get_file_contents '{"owner": "user", "repo": "project", "path": "src/main.ts"}' \
  | jq -r '.content[0].text' > main.ts

6. Aggregate Results from Multiple Servers

{
  mcp-cli call github search_repositories '{"query": "mcp", "per_page": 3}'
  mcp-cli call filesystem list_directory '{"path": "./src"}'
} | jq -s '.'

7. Complex JSON Arguments

For JSON with special characters, use stdin to avoid shell escaping:

# Heredoc (recommended)
mcp-cli call server tool <<EOF
{
  "content": "Text with 'single quotes' and \"double quotes\"",
  "metadata": {
    "nested": "value"
  }
}
EOF

# From file
cat args.json | mcp-cli call server tool

# Using jq to build complex JSON
jq -n '{query: "mcp", filters: ["active", "starred"]}' | mcp-cli call github search

Environment Variables

Example:

# Enable debug mode
export MCP_DEBUG=true
mcp-cli info github

# Use custom config
export MCP_CONFIG_PATH=/path/to/config.json
mcp-cli

# Disable connection pooling
export MCP_NO_DAEMON=true
mcp-cli call server tool '{}'

CLI Options

Example:

# List with descriptions
mcp-cli -d

# Use custom config
mcp-cli -c ./my-config.json info github

# Get help
mcp-cli --help

Common MCP Server Setups

Filesystem Server

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/path/to/directory"
      ],
      "allowedTools": ["read_file", "list_directory", "search_files"],
      "disabledTools": ["delete_file"]
    }
  }
}

Usage:

mcp-cli call filesystem read_file '{"path": "./README.md"}'
mcp-cli call filesystem list_directory '{"path": "."}'

GitHub Server

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ],
      "env": {
        "GITHUB_TOKEN": "${GITHUB_TOKEN}"
      }
    }
  }
}

Usage:

mcp-cli call github search_repositories '{"query": "mcp server"}'
mcp-cli call github get_file_contents '{"owner": "user", "repo": "repo", "path": "README.md"}'

HTTP Server

{
  "mcpServers": {
    "api": {
      "url": "https://mcp.example.com",
      "headers": {
        "Authorization": "Bearer ${API_TOKEN}"
      }
    }
  }
}

Usage:

mcp-cli info api
mcp-cli call api tool_name '{"param": "value"}'

Custom Local Server

{
  "mcpServers": {
    "custom": {
      "command": "node",
      "args": ["./my-server.js"],
      "cwd": "/path/to/project",
      "env": {
        "PORT": "3000",
        "API_KEY": "${API_KEY}"
      }
    }
  }
}

Troubleshooting

Config File Not Found

Error:

Error: No MCP configuration found

Solution:

1. Create mcp_servers.json in current directory or ~/.config/mcp/
2. Or set MCP_CONFIG_PATH environment variable
3. Or use -c flag: mcp-cli -c /path/to/config.json

Server Not Starting

Error:

Error: Failed to connect to server 'github'

Solution:

1. Check command is installed: npx @modelcontextprotocol/server-github --version
2. Verify environment variables are set
3. Enable debug mode: MCP_DEBUG=true mcp-cli info github
4. Check server logs in stderr

Missing Environment Variable

Error:

Error: Environment variable GITHUB_TOKEN not found

Solution:

1. Set the variable: export GITHUB_TOKEN=your_token
2. Or use MCP_STRICT_ENV=false to allow empty values (not recommended)

Tool Not Found

Error:

Error: Tool 'invalid_tool' not found on server 'github'

Solution:

1. List available tools: mcp-cli info github
2. Check for typos in tool name
3. Verify tool isn't disabled in config (disabledTools)

Tool Filtered Out

Error:

Error: Tool 'write_file' is disabled by configuration

Solution:

1. Check config allowedTools and disabledTools
2. Update config to allow the tool
3. Remove or modify filtering rules

JSON Parsing Error

Error:

Error: Invalid JSON arguments

Solution:

1. Use stdin for complex JSON: echo '{}' | mcp-cli call server tool
2. Escape quotes properly: '{"key": "value"}'
3. Use heredoc for multi-line JSON

Connection Timeout

Error:

Error: Request timeout after 1800 seconds

Solution:

1. Increase timeout: export MCP_TIMEOUT=3600
2. Check server responsiveness
3. Disable daemon for fresh connection: export MCP_NO_DAEMON=true

Permission Denied

Error:

Error: EACCES: permission denied

Solution:

1. Check file permissions in cwd directory
2. Verify user has permission to execute command
3. For filesystem server, ensure paths are accessible

AI Agent Integration

System Prompt Template

Add this to your AI agent's system prompt:

## MCP Servers

You have access to MCP servers via the `mcp-cli` CLI.

Commands:
- `mcp-cli info` - List all servers
- `mcp-cli info <server>` - Show server tools  
- `mcp-cli info <server> <tool>` - Get tool schema
- `mcp-cli grep "<pattern>"` - Search tools
- `mcp-cli call <server> <tool> '{}'` - Call with JSON args
- `echo '{}' | mcp-cli call <server> <tool>` - Call from stdin

Workflow:
1. Discover: `mcp-cli info` to see available servers
2. Inspect: `mcp-cli info <server> <tool>` to get schema
3. Execute: `mcp-cli call <server> <tool> '{}'` with arguments

Use stdin for complex JSON to avoid shell escaping issues.

Token-Efficient Workflow

Instead of loading all tool schemas:

# ❌ Avoid: Loading everything into context
mcp-cli -d  # Thousands of tokens

# ✅ Efficient: On-demand loading
mcp-cli                               # List servers (minimal tokens)
mcp-cli info github                   # Show tools when needed
mcp-cli info github search_repositories  # Get schema only when calling
mcp-cli call github search_repositories '{"query": "mcp"}'

Script Generation Example

AI can generate shell scripts combining multiple MCP calls:

#!/bin/bash
# Generated by AI agent to analyze repository

# Search for repos
repos=$(mcp-cli call github search_repositories '{"query": "mcp server", "per_page": 5}')

# Extract URLs
urls=$(echo "$repos" | jq -r '.content[0].text | fromjson | .items[].html_url')

# For each repo, get README
for url in $urls; do
  owner=$(echo "$url" | cut -d'/' -f4)
  repo=$(echo "$url" | cut -d'/' -f5)
  
  echo "=== $owner/$repo ==="
  mcp-cli call github get_file_contents "{\"owner\": \"$owner\", \"repo\": \"$repo\", \"path\": \"README.md\"}" \
    | jq -r '.content[0].text'
done

Best Practices

1. Use stdin for complex JSON - Avoid shell escaping issues
2. Enable descriptions sparingly - Use -d only when needed to save tokens
3. Filter tools in config - Reduce attack surface and simplify discovery
4. Set environment variables - Use ${VAR} substitution for secrets
5. Pipe to jq - Process JSON output efficiently
6. Check schemas first - Use info before call to validate arguments
7. Handle errors - Check exit codes and stderr in scripts
8. Use connection pooling - Default daemon keeps connections warm
9. Search before listing - Use grep for specific tools instead of listing all