owl

OWL-MCP exposes a set of MCP calls for working with OWL ontologies. It maintains an in-memory ontology instance and keeps it in sync with the on-disk file, so edits propagate consistently and can be observed by clients or editors such as Protege. The server focuses on functional-syntax representations of OWL axioms and supports standard CRUD-like operations: creating, reading (finding), updating, and deleting axioms across ontologies. Clients use the MCP protocol to call these functions, receive the current ontology state, and reference the provided file path to locate the ontology on disk. The extension integrates with Goose (or any MCP-enabled AI host) via the uvx-based workflow, enabling AI agents to manage ontologies through simple, string-based function calls.

Prerequisites:

• Python environment (3.8+)
• Access to uv (uvx) tooling and Goose extension setup

1. Install Goose (Desktop or CLI) following the Goose installation guide: https://block.github.io/goose/docs/getting-started/installation/

2. Install the OWL-MCP extension into Goose using UVX:

   • Install directly via URL (if available): (Follow Goose extension installation flow and select OWL-MCP as the extension to install)

   • Or configure manually by adding the extension in Goose and registering the command: uvx owl-mcp

3. Verify installation:

   • Ensure Goose recognizes the OWL-MCP extension and that the uvx command runs with the owl-mcp package.

4. Run Goose with the configured extension and start the MCP workflow. No additional server process beyond Goose + extension is required for this setup.

Note: The exact package name for the uvx-based extension is owl-mcp; if your environment requires a package installation step (e.g., pipx/uvx), ensure the package is accessible in your PATH.

Tips and caveats:

• Ensure the ontology file on disk is accessible and that Goose has permissions to read/write it to keep in-memory and on-disk states synchronized.
• Use the MCP function calls to perform CRUD operations on axioms; each operation updates the in-memory ontology and triggers disk synchronization.
• When displaying OWL in client responses, the server may include human-readable labels for opaque IDs to aid comprehension, following common OBO-style formatting.
• If proté́gé is used alongside, changes to the ontology file on disk will reflect in Protege as long as the file path remains consistent.
• Environment variables are not strictly required by default; if you need to configure file paths or default ontology locations, you can set them via Goose/extension configuration or pass them through MCP calls as needed.