ck:mcp-management

You can manage multiple Model Context Protocol servers by configuring them in a .claude/.mcp.json file and delegating operations to a subagent. This subagent-based architecture keeps the main context clean by loading only necessary tool definitions on demand.

Progressive disclosure for MCP tool discovery works by loading only necessary tool definitions on demand rather than all at once. The skill analyzes which tools are relevant for specific tasks and fetches their definitions progressively to maintain context efficiency.

Yes, you can execute MCP tools with proper parameter handling from a single configuration file. Configure your servers in .claude/.mcp.json and use the execution engine to call MCP tools directly via scripts or subagents.

Yes, the @modelcontextprotocol/sdk is a required dependency for this MCP management approach. It provides the underlying protocol implementation needed to connect to, discover, and execute capabilities from configured servers.

The best way to list all available tools from configured MCP servers is by running the command `npx ts-node scripts/cli.ts list-tools`. This executes the discovery process and outputs the available capabilities from your .claude/.mcp.json configuration.

You should use a subagent-based architecture for Model Context Protocol integrations to achieve context efficiency. By delegating MCP operations to a dedicated mcp-manager subagent, the main context remains clean and only necessary tool definitions are loaded.