@spike-npm-land/spike-mcp-cli

The primary CLI for the spike.land platform. Aggregates multiple MCP servers into a single multiplexed endpoint, with built-in authentication, a server registry, and an interactive chat interface powered by Claude.

Why Lazy Toolset Loading Matters

AI agents work best with small, relevant tool lists. When an LLM receives hundreds of tool definitions at once, it spends context window space parsing tools it will never use — leading to worse decisions and higher cost.

spike-cli solves this with lazy toolset loading: tools are grouped into named toolsets that load on demand. By default, only a handful of gateway tools are visible. When the agent needs chess tools or testing tools, it loads that toolset — and only that toolset enters the context window.

Three benefits:

1. Context savings — hundreds of tools available, but only the active toolset is visible
2. Better AI decisions — a small relevant list beats a giant undifferentiated one
3. One config — any mix of stdio/SSE/HTTP servers, namespaced automatically

Installation

npm install -g @spike-npm-land/spike-mcp-cli

Or run directly from the monorepo:

cd packages/spike-cli
yarn dev

Commands

spike serve

Start the MCP multiplexer server. Connects to all configured MCP servers and exposes their tools through a single endpoint with lazy toolset loading.

spike serve                          # stdio transport (default)
spike serve --transport sse --port 3050  # SSE transport
spike serve --config ./my-config.json    # custom config
spike serve --server myserver="node server.js"  # inline server

Options:

• --config <path> — Path to .mcp.json config file
• --server <name=command> — Add a stdio server inline (repeatable)
• --server-url <name=url> — Add an HTTP/SSE server inline (repeatable)
• --transport <type> — Transport: stdio | http | sse (default: stdio)
• --port <port> — Port for HTTP/SSE transport (default: 3050)
• --namespace-separator <str> — Tool name separator (default: __)
• --no-prefix — Don't namespace tool names

spike chat

Interactive Claude chat session with all configured MCP tools available.

spike chat                           # default model (claude-sonnet-4-6)
spike chat --model claude-opus-4-6   # specify model
spike chat --system "You are a deployment assistant"

Options:

• --model <model> — Claude model to use (default: claude-sonnet-4-6)
• --system <prompt> — System prompt
• --max-turns <n> — Maximum agentic turns per message (default: 20)

Requires CLAUDE_CODE_OAUTH_TOKEN environment variable.

spike shell

Interactive REPL for exploring and calling MCP tools directly.

spike shell
spike shell --config ./my-config.json

spike auth

Manage authentication with spike.land.

spike auth login       # Log in with device code flow
spike auth logout      # Remove stored credentials
spike auth status      # Check current auth status

spike registry

Browse and install MCP servers from the spike.land registry.

spike registry search <query>   # Search for MCP servers
spike registry add <id>         # Add a server to your config

spike alias

Manage command aliases for frequently used operations.

spike alias set <name> <command>   # Create an alias
spike alias remove <name>          # Remove an alias
spike alias list                   # List all aliases

spike completions

Install or remove shell tab completions.

spike completions install     # Install for your shell (bash/zsh/fish)
spike completions uninstall   # Remove completions

spike status

Health check for all configured MCP servers. Reports connection status, tool counts, latency, and environment variable availability.

spike status
spike status --config ./my-config.json

Configuration

spike-cli discovers configuration from .mcp.json files. It searches:

1. Explicit --config path
2. .mcp.json in the current directory
3. ~/.mcp.json in the home directory

Example .mcp.json:

{
  "servers": {
    "my-tools": {
      "command": "node",
      "args": ["./my-mcp-server.js"]
    },
    "remote-server": {
      "url": "https://example.com/mcp"
    }
  }
}

Global Options

• --verbose — Verbose logging to stderr
• --base-url <url> — Base URL for spike.land (default: https://spike.land)
• --help — Show help
• --version — Show version

Development

cd packages/spike-cli

yarn dev          # Run CLI in dev mode (tsx)
yarn build        # Build with tsup
yarn test         # Run tests
yarn typecheck    # Type check

License

BSD-3-Clause