Protégé MCP
Give AI tools live, gated access to the ontology you have open in Protégé Desktop — and chat with an assistant that edits it for you, without ever leaving Protégé.
What is Protégé MCP?
Protégé MCP is a plugin for Protégé Desktop that runs a local MCP (Model Context Protocol) server inside Protégé. It exposes the live, active ontology — the one you have open in the editor — to any MCP-compatible AI client, so the client can explore it and make edits for you through 105 structured tools and 11 guided prompts.
Two ways to use it:
1. As an MCP server for an external client.
Point Claude Code, Codex CLI,
Antigravity, OpenCode,
VS Code, or Claude Desktop at
http://127.0.0.1:8123/mcp and it can read and edit your ontology.
2. As an in-app assistant. The built-in Ontology Assistant chat tab drives a coding-agent CLI you already have (Claude Code, Codex, Antigravity, or OpenCode) back into the same MCP server — so you get a conversational editor inside Protégé, with no separate client to run.
Everything an AI client does flows through Protégé’s own model manager, so:
- Edits appear in the GUI immediately and join the shared Edit ▸ Undo stack.
- Safety gates apply to every write — a read-only mode and a confirm-each-write prompt you control from Settings ▸ MCP.
- The server is local and authorized — it binds to
127.0.0.1unless you deliberately pick another interface, and requires an OAuth authorization or a bearer token on every request.
Why use it?
- Author faster in natural language. “Create a class Widget under Product with a definition and label,” or “find every class with no definition and propose fixes” — the assistant turns that into concrete, previewable axiom changes.
- Stay in control. Preview edits before they apply, gate or confirm every write, and undo anything with one keystroke. The AI can never escalate past the gates you set.
- Reason and query. Run the reasoner, explain entailments, find unsatisfiable classes, and run SPARQL 1.1 over the active ontology and its imports — all through tools.
- Reuse and materialize with evidence. Govern SSSOM mappings, inspect OLS4 or experimental OntoPortal terms through explicit BioPortal, AgroPortal, or custom compatible endpoint configuration and reuse proposals, validate SWRL rules against an exact reasoner profile, preview bounded inferences, and run cancellable owner-scoped jobs where the live plugin supports them.
- Govern your project. Audit modelling quality and project-governance policy (OWL 2 profile conformance, IRI policy, required annotations, module ownership / import layering), load persisted invariants/CQs/SHACL from a checked-in policy, and distinguish a validation failure from a check that could not run.
- Enforce project boundaries. Policy-relative paths stay below the canonical project root, remote documents/imports require explicit network policy and capability, and locked import content is checked automatically before a project gate or change-set commit can pass.
- Exchange a portable project identity. Validate RO-Crate metadata and compute a W3C RDFC-1.0 identity for the asserted root RDF dataset, while keeping dependency checksums in an import lock.
- No API key stored by Protégé. The Ontology Assistant reuses your existing CLI login; nothing is kept in the plugin.
Quick start
- Install the plugin (drop the jar into Protégé’s
plugins/folder, or use File ▸ Check for plugins). Protégé must run on Java 17+. - Open the MCP Server view, start the server, and note the URL (
http://127.0.0.1:8123/mcp). - Either connect an external client or open the Ontology Assistant tab and type a request.
Documentation map
| Section | What’s inside |
|---|---|
| Installation | Requirements (Java 17+), manual install, and Check for plugins. |
| Connecting a client | The server model (ports, OAuth vs. bearer token) and per-environment recipes for Claude Code, Codex CLI, Antigravity, OpenCode, VS Code, and Claude Desktop. |
| Ontology Assistant | The in-Protégé chat: what it is, how it works, attachments, privacy, and settings. |
| Tools | All 105 tools by category. |
| Prompts | The 11 guided workflows available to MCP clients. |
| Headless CLI | Run policy validation, full project QC, releases, and asserted semantic diff without Protégé. |
| Ontology CI | Gate your own ontology repository’s PRs with the reusable, fork-safe workflow: full project QC with JSON/JUnit/SARIF evidence, a dry-run release preview, and the asserted diff. |
| Project policy & QC | Policy v1/v2/v3 discovery and validation, workspace membership, SSSOM/provider/job/materialization controls, fingerprints, persisted QC assets, examples, and strict gate semantics. |
| RO-Crate & RDFC | Package portable project metadata and produce a canonical RDF dataset identity. |
| Commercial platforms | The supported exchange boundary, product-specific caveats, and requirements for future connectors. |
| Contributing | Build from source, run the tests, project layout, and how to add a tool. |
| Performance tests | Versioned representative fixtures, regression budgets, and the opt-in benchmark command. |
| Live integration test | Automated real-Protégé release gate and the platform packaging checklist. |
| Versioning & releases | The SemVer + Keep-a-Changelog scheme and how releases are cut. |
| Changelog | Release notes for every version. |
Protégé MCP is licensed under the BSD 2-Clause License.