mcpcute

You can manage multiple MCP servers behind a single, lightweight interface. This MCP aggregates several MCP endpoints so you can discover, inspect, and execute tools without loading every tool’s full schema upfront. It starts instantly and fetches tool details on demand, helping you keep AI agent context clean and focused.

How to use

You interact with seven core operations, divided into two levels. Use MCP-level operations to discover and inspect MCP servers, then switch to tool-level operations to inspect and run tools from a chosen MCP. The workflow emphasizes lazy loading: you discover MCPs quickly, inspect details on demand, and execute tools when needed.

How to install

Prerequisites: you need Node.js (LTS version) and a modern shell. You will install the MCP aggregator globally so you can run its commands from anywhere.

Install the package globally using your preferred package manager.

Configuration

Configure the MCP aggregator to point to your MCP servers. You’ll provide a JSON file that lists each MCP with its command and arguments. The aggregator supports on-demand connections, so it will only connect to an MCP when you request details or when you execute a tool.

Create a configuration file named mcpcute.config.json in your working directory and define the MCP servers you want to aggregate. The example below shows two MCP servers: a filesystem-based MCP and a fetch-based MCP.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
    },
    "fetch": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-fetch"]
    }
  }
}

Security and secrets

Only expose the MCP interfaces you need. Do not include sensitive credentials in your configuration. If you must pass secrets, use environment variables or a secret manager supported by your runtime.

Cache and performance

Tool discovery results are cached locally to avoid reconnecting to every MCP on subsequent runs. The cache location can be overridden with an environment variable.

Troubleshooting

If an MCP does not respond, ensure the command and arguments in your config are correct and that the MCP server is reachable. Remember that the aggregator connects on-demand, so you won’t see a large upfront startup delay.

Notes and examples

Common workflow patterns are described below to help you discover and use tools efficiently.

Discover MCPs without connecting: use list_mcps() or search_mcps("term") to locate MCPs by name.

Examples of typical commands

# Discover all configured MCPs (no connections opened yet)
list_mcps()

# Inspect a specific MCP's details and its tools (on-demand connection)
get_mcp_details("fetch")
list_tools("fetch")

# Get details for a specific tool and then execute it
get_tool_details("read_file")
execute_tool({ mcp_name: "filesystem", tool_name: "read_file", arguments: { path: "/tmp/example.txt" } })

Available tools

list_mcps

List all available MCP servers with their connection status to help you see what is configured and ready.

search_mcps

Search for MCP servers by name to quickly locate the ones you want to inspect or use.

get_mcp_details

Get detailed information about an MCP, including its available tools and capabilities.

list_tools

List all tools for a specific MCP to understand what actions you can perform.

search_tools

Search for tools across MCPs, or scope the search to a single MCP to narrow results.

get_tool_details

Get the full schema and description for a tool so you know required arguments and outputs.

execute_tool

Execute a tool from a chosen MCP by providing the MCP name, tool name, and required arguments.