Why can't a human use MCP too?

Command-line interface for any Model Context Protocol (MCP) server.

MCP enables AI assistants to interact with tools and data sources. mcp-cli brings that same power to the command line, making MCP accessible to humans for:

• 🤖 Shell scripting and automation
• 🔧 DevOps and CI/CD pipelines
• 🧪 Quick testing of MCP servers
• 📊 Data operations and queries
• ⚡ Rapid prototyping

• 🔌 Universal: Works with ANY MCP server
• 🚀 Zero Schema Config: Discovers tools dynamically from the server
• 🎯 Automatic CLI Generation: Tool schemas → Commander.js options (magically!)
• 📝 Clean Output: Parseable results perfect for piping
• 🛠️ Human-Friendly: No JSON-RPC knowledge needed

npm install -g mcp-cli

# Use any MCP server directly
mcp-cli --server "uvx chroma-mcp" --list-tools

# Call a tool
mcp-cli --server "uvx chroma-mcp" chroma_list_collections

Create ~/.mcp-cli.json:

{
  "servers": {
    "chroma": {
      "command": "uvx",
      "args": ["chroma-mcp", "--data-dir", "~/.chroma"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "your-token"
      }
    }
  },
  "defaultServer": "chroma"
}

Then use named servers:

# List tools from named server
mcp-cli --use chroma --list-tools

# Use default server (no --use needed)
mcp-cli chroma_query_documents --help

# Call a tool
mcp-cli --use github github_create_issue --repo "owner/repo" --title "Bug"

mcp-cli --server "uvx chroma-mcp" --list-tools

The CLI calls the MCP tools/list method - the server returns:

{
  "tools": [
    {
      "name": "chroma_query_documents",
      "description": "Query documents with semantic search",
      "inputSchema": {
        "type": "object",
        "properties": {
          "collection_name": { "type": "string" },
          "query_texts": { "type": "array" },
          "n_results": { "type": "integer" }
        },
        "required": ["collection_name", "query_texts"]
      }
    }
  ]
}

mcp-cli chroma_query_documents --help

Output:

Usage: mcp-cli chroma_query_documents [options]

Query documents with semantic search

Options:
  --collection_name <string>  Collection name (required)
  --query_texts <json>        Query texts (JSON array) (required)
  --n_results [integer]       Number of results
  -h, --help                  display help for command

mcp-cli chroma_query_documents \
  --collection_name memories \
  --query_texts '["search term"]' \
  --n_results 5

All generated from the MCP schema - zero manual configuration!

# List collections
mcp-cli --use chroma chroma_list_collections

# Query documents
mcp-cli chroma_query_documents \
  --collection_name memories \
  --query_texts '["important notes"]' \
  --n_results 10

# Add documents
mcp-cli chroma_add_documents \
  --collection_name memories \
  --documents '["Note 1", "Note 2"]' \
  --ids '["id1", "id2"]'

# Create issue
mcp-cli --use github github_create_issue \
  --repo "myorg/myrepo" \
  --title "Bug: Login fails" \
  --body "Description here"

# List pull requests
mcp-cli --use github github_list_pull_requests \
  --repo "myorg/myrepo" \
  --state open

#!/bin/bash
# Query memories and create GitHub issues for bugs

# Get bug reports from vector db
BUGS=$(mcp-cli chroma_query_documents \
  --collection_name bugs \
  --query_texts '["high priority bugs"]' \
  --n_results 5)

# Parse and create issues
echo "$BUGS" | jq -r '.documents[0][]' | while read bug; do
  mcp-cli --use github github_create_issue \
    --repo "myorg/myrepo" \
    --title "$bug"
done

The CLI looks for config in this order:

1. ./.mcp-cli.json (current directory)
2. ~/.mcp-cli.json (home directory)
3. ~/.config/mcp-cli/config.json

{
  "servers": {
    "server-name": {
      "command": "executable",
      "args": ["arg1", "arg2"],
      "env": {
        "ENV_VAR": "value"
      },
      "description": "Optional description"
    }
  },
  "defaultServer": "server-name"
}

Note: The config ONLY specifies HOW to connect to servers. Tool schemas are discovered dynamically from the MCP server itself.

# Specify server inline
mcp-cli --server "command args..." <tool-name> [options]

# Use named server from config
mcp-cli --use <server-name> <tool-name> [options]

# Use default server from config
mcp-cli <tool-name> [options]

# List available tools
mcp-cli --list-tools

# Get help for a tool (dynamically generated!)
mcp-cli <tool-name> --help

# Version
mcp-cli --version

Results are automatically formatted for readability:

# Simple values
$ mcp-cli chroma_get_collection_count --collection_name memories
42

# JSON objects (pretty-printed)
$ mcp-cli chroma_query_documents --collection_name memories --query_texts '["test"]'
{
  "ids": [["doc1", "doc2"]],
  "documents": [["First doc", "Second doc"]],
  "distances": [[0.1, 0.2]]
}

Perfect for piping to jq:

mcp-cli chroma_query_documents ... | jq '.documents[0][0]'

MCP servers expose powerful capabilities through a standard protocol. But to use them, you typically need to:

1. Write custom client code
2. Handle JSON-RPC manually
3. Parse tool schemas yourself
4. Build your own CLI

mcp-cli does all of this automatically.

You just specify the server command, and the CLI:

• ✅ Connects via stdio
• ✅ Discovers available tools
• ✅ Generates CLI commands from schemas
• ✅ Handles all the protocol details
• ✅ Formats output nicely

Works with any MCP server that implements the standard protocol. Examples:

• uvx chroma-mcp - ChromaDB vector database
• npx @modelcontextprotocol/server-github - GitHub API
• npx @modelcontextprotocol/server-filesystem - Filesystem operations
• npx @modelcontextprotocol/server-postgres - PostgreSQL database
• Custom MCP servers you build

# Clone repo
git clone https://github.com/YOUR_USERNAME/mcp-cli.git
cd mcp-cli

# Install dependencies
npm install

# Build
npm run build

# Test locally
node bin/mcp-cli.js --server "uvx chroma-mcp" --list-tools

Contributions welcome! Please open an issue or PR.

MIT

"Why can't a human use MCP too?"

Because now they can. 🚀