Ontology-driven skill compiler for deterministic agent tool-calling in MCP environments
Ontoskills by Marea Software is an MCP server and compiler that enforces deterministic, ontology-driven skill management for agentic workflows. It converts markdown skill descriptions into machine-usable ontologies and exposes a queryable registry so agents locate precise capabilities instead of relying on probabilistic guesses. The package includes an OWL-based compiler, a runtime built in Rust for low-latency responses, and SHACL validation to gate unsafe or malformed skills. AI engineers and teams building multi-agent systems benefit from predictable tool-calling and reduced token waste when integrating the server into MCP-compatible runtimes.
What tasks can you actually use it for?
The tool targets deterministic capability management inside production multi-agent systems and fail-closed architectures, where unpredictable tool calls are unacceptable. It supports semantic orchestration across domain-agnostic agents and bridges human-readable documentation with machine-executable representations, which reduces token waste by narrowing the agents' search space. Teams building reliable agent chains adopt it to make capability discovery auditable and to enforce predictable runtime behavior under formal control.
How accurate are the discovery and validation steps?
Discovery uses structured queries and validation, applying SPARQL-based lookup to locate exact matches in the capability graph. A SHACL gatekeeper validates compiled entities before they are callable, enforcing structural constraints and safety shapes. That design reduces hallucinated tool calls and produces deterministic audit trails for decision points, while the ultimate correctness of mappings depends on the precision of authored documentation.
What file formats does it support and where does it run?
Input is standard SKILL.md documentation, which the compiler transforms into validated RDF/Turtle ontologies for machine consumption. The server conforms to the Model Context Protocol and interoperates with MCP-compatible hosts. Typical integrations listed include:
- Claude Desktop
- Cursor
- Windsurf
Is it straightforward to adopt for development teams?
The developer focuses on deterministic AI and semantic web technologies, so adoption favors teams with ontology and verification experience. Integration requires authorship workflows for ontology artifacts and alignment with MCP endpoints. Teams already using formal verification or ontology pipelines gain auditability and predictable tool-calling; teams without semantic-web expertise encounter a nontrivial learning curve and must add documentation-to-ontology practices before full production rollout.
Best fit: protocol-aligned engineering teams that accept an ontology learning curve
The tool is a disciplined option for AI engineers and enterprise teams that require verifiable, deterministic capability selection within MCP-based environments. It demands semantic-web expertise and protocol-aligned workflows, which narrows suitability to teams prepared to manage ontology artifacts. Plan a phased rollout: validate a small set of skills, include ontology checks in CI, and treat skill authoring as part of engineering responsibilities.
