Show HN: A better alternative to CLI and MCP for local tools (github.com)

A new library has been introduced that serves as a more efficient alternative to CLI and MCP for local tools, leveraging named pipes as the communication layer for agentic tool servers. This architecture allows persistent background processes to provide functionalities like LLM inference, text-to-speech, vector search, and browser automation to a local Python orchestrator. Because named pipes transfer data through kernel memory instead of a network stack, they significantly reduce latency and complexity, making them ideal for real-time applications such as voice assistants. Tool servers remain resident between calls, retaining resources in memory and eliminating the need for repeated data loading. This development holds particular significance for the AI/ML community as it streamlines the operation of local AI tools, enhancing performance by removing unnecessary overhead from process-spawning and state-reloading. The library includes command-line utilities for ad-hoc interactions with running servers, and can be integrated seamlessly with agentic coding tools like Claude Code. Technical implementations feature hierarchical abstractions for managing data and message flow, allowing for the efficient handling of concurrent clients. With support for advanced features like in-memory state retention and real-time capability execution, this innovation paves the way for more agile and responsive AI applications.