Releases·Tool Release
KitOps (CNCF Sandbox) versions MCP servers, agent skills, and models as OCI ModelKits pushed to existing registries
A practitioner describes using KitOps (CNCF Sandbox) to package models, datasets, MCP server bundles, agent skills, and policies as versioned OCI artifacts called ModelKits, solving reproducibility problems where MCP configs, skills, and prompts lived in separate places. The CLI mimics Docker workflows and works with existing OCI registries.
1 engagement1 sourceMon, Sep 21, 2026, 12:29 PM
AI Overview
- KitOps packs models, datasets, MCP server bundles, agent skills, and policies into versioned OCI artifacts called ModelKits via Docker-like CLI commands `kit pack`/`push`/`pull`/`unpack`
- Kitfile supports `mcpServers` entries for MCP bundles (`.mcpb`) as of v1.15, and optional CNCF ModelPack output via `kit pack . --use-model-pack`
- Selective unpack avoids downloading full artifacts; KitOps complements containers rather than replacing Docker
Related
Products·launched
Pullrun: lightweight runtime runs OCI images as containers or microVMs with MCP for AI agents
Pullrun is a ~20 MB runtime that executes OCI images as either containers (via runc) or microVMs (Firecracker on Linux, Apple Silicon VMs on macOS) from the same image, eliminating the need for a separate VM build step. It integrates natively with the Model Context Protocol (MCP), exposing 15 runtime operations as MCP tools for AI agents like Claude Code and opencode. This gives agents a consistent, sandboxed execution environment without managing Docker or full VMs.
Pullrun is a ~20 MB runtime that executes OCI images as either containers (via runc) or microVMs (Firecracker on Linux, Apple Silicon VMs on macOS) from the same image, eliminating the need for a separate VM build step. It integrates natively with the Model Context Protocol (MCP), exposing 15 runtime operations as MCP tools for AI agents like Claude Code and opencode. This gives agents a consistent, sandboxed execution environment without managing Docker or full VMs.
3 engagement1 sourceredditTue, Jul 28, 2026, 09:10 PM
Releases·Tool Release
Developer releases Trivium CLI tool for reproducible agent skill management
A developer released Trivium, a CLI tool that pins agent skills to exact Git commits and records them in a `skills.lock` file, enabling reproducible skill environments. The tool supports named environments for switching between different skill configurations, addressing the common problem of tracking skill revisions during evaluation.
A developer released Trivium, a CLI tool that pins agent skills to exact Git commits and records them in a `skills.lock` file, enabling reproducible skill environments. The tool supports named environments for switching between different skill configurations, addressing the common problem of tracking skill revisions during evaluation.
2 engagement1 sourceredditSat, Jul 18, 2026, 07:26 PM
Products·launched
Tachyon: MCP server framework for Java and Kotlin
Tachyon is a framework for building Model Context Protocol (MCP) servers in Java 21+ and Kotlin without Spring or reactive APIs. It leverages Netty, Streamable HTTP, virtual threads, and Kotlin coroutines. It is currently in 1.0 beta and aims to simplify MCP server development for JVM developers.
Tachyon is a framework for building Model Context Protocol (MCP) servers in Java 21+ and Kotlin without Spring or reactive APIs. It leverages Netty, Streamable HTTP, virtual threads, and Kotlin coroutines. It is currently in 1.0 beta and aims to simplify MCP server development for JVM developers.
1 engagement1 sourcehackernewsTue, Aug 11, 2026, 01:01 PM
Products·launched
SkillRanger: package manager for AI agent skills with checksums and audits
SkillRanger is a package manager for AI agent skills, treating skills like packages with versions, checksums, and static audits. It scans a repository, detects the stack, and recommends a small set of compatible skills, avoiding clutter. It solves the problem of trusting and managing skill files in projects using Codex or Claude Code.
SkillRanger is a package manager for AI agent skills, treating skills like packages with versions, checksums, and static audits. It scans a repository, detects the stack, and recommends a small set of compatible skills, avoiding clutter. It solves the problem of trusting and managing skill files in projects using Codex or Claude Code.
3 engagement1 sourceredditSun, Aug 9, 2026, 01:09 PM
Social·Community
Agent Plugins standard packages Skills and MCP servers into portable folders
A proposed standard called Agent Plugins packages Agent Skills and MCP servers into a single portable folder that any compatible agent client can load. The approach reuses existing standards rather than reinventing them, and isolates failures so a broken MCP server doesn't block valid Skills. It aims to keep the core small while allowing client-specific extensions.
A proposed standard called Agent Plugins packages Agent Skills and MCP servers into a single portable folder that any compatible agent client can load. The approach reuses existing standards rather than reinventing them, and isolates failures so a broken MCP server doesn't block valid Skills. It aims to keep the core small while allowing client-specific extensions.
5 engagement1 sourceredditWed, Aug 19, 2026, 01:53 PM