Install skills, enable them locally, and query them through MCP.
OntoSkills is a minimal product surface for deterministic agent skills: one CLI, one local runtime, one official registry, one optional compiler when you need custom source skills.
npx ontoskills install mcp `ontoskills` installs components, pulls registry skills, and manages local state.
`ontomcp` exposes enabled ontologies through a compact, package-aware tool surface.
`ontocore` remains optional for teams that author or import raw `SKILL.md` sources.
Install a skill from the official registry
Search all verified skills published in the official registry and copy the exact install command for your local runtime.
The Distribution Problem
Skills are useful only when they are discoverable, versioned, and runnable. Raw files, compiled ontologies, and runtime state drift apart quickly without a single source of truth.
- ✗ Fragmented installs — runtime, registry, and compiler live in different places
- ✗ No provenance — users cannot easily tell where a skill came from or how it was built
- ✗ Ad-hoc execution — source skills, compiled skills, and runtime state need different handling
The OntoSkills Stack
A single product surface wraps the official registry, the managed MCP runtime, and the optional compiler. Runtime-only users install and query; authors can additionally compile custom skills locally.
- ✓ Managed CLI —
ontoskillsinstalls, updates, enables, and removes components - ✓ Official registry — compiled skills are published once and pulled by qualified id
- ✓ Runtime + compiler —
ontomcpserves the catalog;ontocorecompiles custom sources
Key Capabilities
Everything you need to ship deterministic, queryable skill ontologies without gluing multiple tools together.
Official Registry
Published compiled skills are discoverable by qualified id and versioned in a public registry.
Managed CLI
The ontoskills command installs, updates, enables, disables, and removes product components.
Remote Install
Install a single compiled skill from the registry and keep it disabled until you enable it.
Runtime MCP
OntoMCP loads the enabled ontology set and serves package-aware context through a small tool surface.
Compiler Workflow
OntoCore turns raw SKILL.md sources into validated ontologies when you need custom skills.
Deterministic Resolution
Qualified ids, provenance, and explicit enable/disable state keep lookups predictable.
How It Works
From registry entry to enabled skill in one managed flow.
Registry Layer
The official registry publishes compiled skills and metadata that the CLI can install remotely.
Runtime Layer
OntoMCP loads enabled ontologies from the managed home directory and answers structured tool calls.
Compiler Layer
OntoCore compiles custom SKILL.md sources when you need to author or extend the registry locally.
How It Works
From registry discovery to local compilation in five steps.
Getting Started
Three paths cover the full product: runtime only, registry installs, or custom compilation.
Install OntoMCP
npx ontoskills install mcp
Install a registry skill
npx ontoskills search hello npx ontoskills install marea.greeting/hello npx ontoskills enable marea.greeting/hello
Install OntoCore
npx ontoskills install core ontoskills init-core ontoskills compile
Roadmap
This evolves as we build.
OntoCore
Source skills compile into validated ontologies and reusable knowledge nodes
OntoMCP
Package-aware MCP runtime serves enabled ontologies through a small tool surface
OntoSkills
Managed CLI orchestrates installs, updates, enable/disable, and local home layout
OntoSkills Registry
Official registry publishes compiled skills and supports remote installs by qualified id
Community Imports
Third-party registries, source repository imports, and stricter release/update automation
Track progress on GitHub and the official registry at ontoskills-registry
Install. Enable. Query.
Start with the managed CLI, pull from the official registry, and let OntoMCP serve the enabled skill set. Install OntoCore only when you need to author or import custom source skills.
npx ontoskills install marea.greeting/hello