vc:mcp-management

You can orchestrate multiple MCP servers by loading configurations from .claude/.mcp.json, connecting via the MCP client, and using a dedicated CLI to manage tools. This isolates tool discovery and execution, keeping the main context clean and focused on the primary task.

Intelligent tool discovery analyzes available MCP tools, prompts, and resources across connected servers to determine which are relevant for a specific task. It progressively loads only necessary tool definitions on-demand to minimize context usage and improve execution efficiency.

You execute tool calls across MCP server endpoints using the provided execution engine and CLI scripts. The system routes calls to the correct server with proper parameter handling, ensuring context-efficient delegation when invoking external tools.

Yes, MCP orchestration works with the @modelcontextprotocol/sdk to connect to and manage multiple MCP servers. It requires loading server configurations from .claude/.mcp.json and uses the SDK client to establish connections for tool discovery and execution.

Loading all MCP tool definitions at once consumes excessive context because each server exposes numerous tools, prompts, and resources. On-demand loading progressively imports only the necessary definitions, preventing context bloat and keeping the AI focused.

Yes, you can use the MCP management CLI scripts to list available tools from all configured servers. Running the CLI connects to the servers defined in .mcp.json and aggregates the discovered tools, prompts, and resources for analysis and selection.