Features

Everything your agents need, in one service

Skillberry Store manages three resource types — tools, skills, and snippets — and exposes them through the interfaces agents actually use: REST, MCP, and mountable filesystems.

The building blocks

Tools, skills & snippets

Tools

Executable units for agentic workloads — a Python module or an MCP-backed tool. Each has state, visibility, tags, a version, and a packaging format, and runs with parameters inside a sandbox.

Skills

Reusable groupings of tools and snippets, compatible with the Anthropic skill format (SKILL.md + tools + docs). Import, export, and detect Anthropic skills directly.

Snippets

Code fragments and documentation stored with syntax highlighting and searchable by content or tags — reference material and shared utilities your tools depend on.

Discover

Semantic & classic search

Shortlist tools, skills, and snippets by meaning or by keyword. Every resource is embedded (faiss by default, 384 dimensions) so agents and humans find the right capability fast.

  • Search across tools, skills, snippets, VMCP and vNFS servers
  • Filter and label with namespaces for categorization
  • Lifecycle controls for state and visibility
Tools registry with search in the Skillberry Store UI
Run safely

Sandboxed execution

Invoke tools with parameters inside a Docker (or Podman) sandbox, so agent-generated code never touches your host. Dependencies can be auto-detected, and execution is available from the UI, the REST API, the CLI, and over MCP.

  • Isolated, parameterized tool execution
  • Automatic tool dependency detection
  • Optional local execution mode for development
Tool execution dialog in the Skillberry Store UI
Expose · MCP

Virtual MCP servers (VMCP)

A VMCP server exposes a single skill's tools and snippets as a standalone MCP endpoint on its own port. Point Claude, Cursor, or any MCP client at it — or expose every REST operation as MCP tools through the control API at /control_sse.

  • One skill → one standalone MCP endpoint
  • MCP frontend: serve your tools to any agent framework
  • MCP backend: consume and route tools from external MCP servers
Virtual MCP servers in the Skillberry Store UI
Expose · Filesystem

Mountable vNFS filesystems

A vNFS server exposes a single skill as a read-only filesystem over WebDAV or NFSv3. Any tool that can mount a network drive — including Claude Code — reads skill files directly via mount or rclone, no REST API required.

  • WebDAV (default) or NFSv3 protocols
  • Mount with rclone, davfs2, or native NFS
  • The UI shows pre-filled mount commands per server
Virtual NFS servers in the Skillberry Store UI
Operate

Observability & persistence

Understand how your tools are used and where your data lives. SBS emits Prometheus metrics and OpenTelemetry traces, and persists resources to the filesystem or to GitHub repositories via version-controlled hooks.

  • Prometheus metrics (prefixed SBS_) and Jaeger-compatible traces
  • Filesystem or GitHub-backed persistence
  • Pluggable vector DB backends: faiss, Chroma, LanceDB
Observability view in the Skillberry Store UI
Interfaces

Reach the store however you work

InterfaceWhat it's for
Web UIVisual management of tools, skills, snippets, VMCP & vNFS servers, and plugins. React + PatternFly, with real-time updates.
REST / OpenAPIFastAPI endpoints with Swagger docs at /docs, using tools-manifest artifacts.
CLIThe sbs command — every API operation from your terminal. See the CLI guide →
MCP control APIEach REST operation exposed as an MCP tool at /control_sse for agentic workflows.
VMCPPer-skill virtual MCP servers on dedicated ports.
vNFSPer-skill mountable WebDAV / NFSv3 filesystems.
Python SDKConsume the service programmatically via the skillberry-store SDK (also ships the CLI).

Want to extend it? The plugin ecosystem adds AI-powered creation, evaluation, security scanning, import, and optimization — without touching core code.