Skillgesture Provides On-Demand Skills to AI Agents Through a Local MCP Server

Summary

Skillgesture is a local Model Context Protocol (MCP) server for organizing and delivering reusable skills to AI agents. It stores skills as Markdown in a central repository under ~/.skillgesture, with groups, skills, and subskills forming the catalog hierarchy. Catalog data, folder associations, and agent sessions are persisted separately as JSON, while the repository keeps immutable Markdown versions and points to an active version for consistent reads. Agents create or resume durable sessions, configure the folders they are working with, and receive the deduplicated union of global and exactly path-associated skills. The skill_tree tool returns a lightweight metadata index without Markdown content, while skill_read loads one active skill or subskill only when requested and can separately retrieve referenced resources. The skill_manage tool supports catalog changes, session operations, associations, and enabling or disabling nodes. Multiple folders can be loaded at once, but associations apply only to exact canonical paths rather than descendant directories. The server uses MCP stdio transport, keeps diagnostics on stderr, and requires Node.js 24 or later and npm 12 or later. Atomic writes, a shared cross-process lock, and expected-version or revision checks are intended to prevent conflicting updates when multiple MCP processes share the repository.