Protege

hakjuoh3stars

An MCP (Model Context Protocol) server for Protégé Desktop: exposes the live, active ontology to MCP clients (Claude Code, Codex, …) and edits it through Protégé's shared OWLModelManager — plus an in-

Install Run on PlugMCP

GitHub Website

Run on PlugMCP

1# Protege

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: BSD-2-Clause
Last checked: 2026-10-28

Stars: 3
Tools: —
Uptime: —

Overview

About Protege

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Protege?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Protege need an API key?

No required environment variable is listed on this record.

What transport does Protege use?

Protege runs locally over stdio.

When was Protege last checked?

Last checked 2026-10-28. An MCP (Model Context Protocol) server for Protégé Desktop: exposes the live, active ontology to MCP clients (Claude Code, Codex, …) and edits it through Protégé's shared OWLModelManager — plus an in-Protégé AI chat assistant.

From the publisher

Documentation

Protégé MCP

Protégé MCP is a plugin for Protégé Desktop that runs a local MCP (Model Context Protocol) server. It gives MCP-compatible AI tools — such as Claude Code, Codex, Antigravity, and OpenCode — live access to the ontology you have open in Protégé, so they can explore it and make edits for you. It also ships an in-app Ontology Assistant chat that drives your own claude / codex / agy / opencode CLI against the same server.

📖 Documentation

Guide	What's inside
Installation	Requirements (Java 17+), manual install, and Check for plugins.
Connecting a client	The server model (ports, OAuth vs. token) + Claude Code, Codex CLI, VS Code, Claude Desktop recipes.
Ontology Assistant	The in-Protégé chat: what it is, attachments, privacy, settings.
Tools	All 105 tools by category.
Prompts	The 11 guided workflows available to MCP clients.
Project policy & QC	Policy v1/v2/v3 discovery and validation, workspace membership, 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, and 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 · Changelog	The release scheme and notes for every version.

Highlights

105 structured tools + 11 guided prompts over the live, active ontology — explore, edit, curate, govern, extract modules, run SWRL rules, SPARQL and SHACL, and reason.
0.8.0 release capabilities — governed SSSOM mappings, external-term reuse proposals, exact reasoner capability reports, preview-first inference materialization, and bounded asynchronous jobs.
Enforced project boundaries (0.6.1) — direct paths are confined to the canonical project root with request-scoped filesystem capabilities, policy network/import controls block unapproved fetches, locked-import checksums are automatic gates, and verified rollback rejects a failing delta before it reaches the live ontology.
Least-privilege OAuth — all 105 tools declare and enforce ontology, release, filesystem, network, and server capabilities from the propagated principal. Explicit read grants cannot mutate, release, or open caller-selected files; existing mcp grants and the static token retain the documented local-admin compatibility profile.
Attributable local operations — every authorized, denied, successful, and failed tool call is written to a rotated owner-only stream without tokens, prompts, attachments, or ontology content; export_audit_log produces an explicitly confirmed, project-contained review artifact.
One endpoint, any number of windows (0.5.0) — a shared broker owns the configured port across every Protégé window and instance and routes each MCP session to a live window, so one fixed URL (http://127.0.0.1:8123/mcp) always works.
Safe, testable authoring (0.4.0, strengthened in 0.6.1) — apply_changes verify=report|rollback runs the isolated project/change-set gate before commit; a re-runnable competency-question suite and SPARQL invariants (verify_ontology) catch regressions; a single run_qc_suite gate rolls up reasoner + profile + structural + invariants + CQ checks; and search_entities ranks policy-configured RDFS/SKOS/OBO preferred labels and synonyms, explains each match/collision, and keeps fuzzy/synonym reuse review-only so an assistant grounds a term before minting.
Reproducible project QC (0.6.0) — discover and validate a checked-in policy, fingerprint the active ontology and loaded imports, resolve persisted invariant/CQ/multi-SHACL assets inside a confined project root, and run reasoner/profile/structural/governance/invariant/CQ/SHACL checks over one isolated snapshot with distinct pass / policy fail / execution error outcomes. The selected reasoner's exact Protégé configuration is preserved without classifying or mutating the live reasoner.
Portable project identity (0.6.0) — validate RO-Crate 1.0–1.3 metadata and report a W3C RDFC-1.0 + SHA-256 identity for the asserted root RDF dataset, separately from the editor revision fingerprint and imported-artifact checksums.
Import integrity inspection (0.6.0) — inspect deterministic direct/transitive import graphs, missing documents, cycles, source locations, and version conflicts; document loads keep the compatible warning default and support strict missing_imports=error for project/release workflows.
Edits are GUI-visible and undoable — every edit goes through Protégé's shared OWLModelManager, so it appears in the editor immediately and joins the Edit ▸ Undo stack.
Safe by default — a local server, bound to loopback (127.0.0.1) unless you change the bind-address preference, that requires OAuth or a bearer token on every request, with optional read-only mode and confirm-each-write.
In-app Ontology Assistant — chat that drives your existing claude, codex, agy, or opencode CLI back into the server with one short-lived, attributable credential per turn instead of the static admin token; no API key is stored by Protégé.

Requirements

Protégé must run on Java 17+ — the bundle is Java 17 bytecode and its OSGi manifest requires JavaSE 17, so it will not load on Java 11. If Protégé ships a Java 11 runtime, point it at a 17+ JDK via the PROTEGE_JAVA_HOME environment variable. See Installation → Requirements.

Install

Download protege-mcp-<version>.jar from the latest release.
Drop it into Protégé's plugins/ folder (keep only one such jar — the bundle is a singleton).
Restart Protégé on a Java 17+ JVM.

Prefer automatic discovery and updates? Use Protégé's File ▸ Check for plugins. Full instructions, including per-OS plugin paths, are in the Installation guide.

Quick start

Settings ▸ MCP — configure and start the server (default http://127.0.0.1:8123/mcp). The MCP Server view shows the bound URL and bearer token.
Either connect an MCP client (Claude Code, Codex CLI, Antigravity, OpenCode, VS Code, Claude Desktop) or open the Ontology Assistant tab and type a request — "What classes are in this ontology?", then "Create a class FooBar under Thing."

Building from source

Requires Git, Apache Maven, and a JDK 17+:

git clone https://github.com/hakjuoh/protege-mcp.git
cd protege-mcp
mvn clean package

The reactor produces plugin/target/protege-mcp-<version>.jar (the self-contained OSGi plugin) and cli/target/protege-mcp-cli-<version>-all.jar (an executable headless Java 17 CLI). The CLI supports policy/full-QC validation, deterministic import locks, verified releases, asserted/manifest-backed diff, and a bounded project-confined MCP server via serve --transport stdio --project FILE, all without a Protégé installation. A fork-safe reusable CI pair applies the base branch's trusted policy to candidate ontology bytes and preserves JSON/JUnit/SARIF/release-checksum evidence. See Ontology CI and the Contributing guide for project layout and how to add a tool.

License

The software is licensed under the BSD 2-Clause License.

Protege

hakjuoh3stars

Install Run on PlugMCP

GitHub Website

Run on PlugMCP

1# Protege

README.mdMCP server

Transport: stdio
Status: Local

Power your agent

Add this server

Install from the publisher. Not a PlugMCP endpoint.

Run on PlugMCP

Host a copy with our keys and logs.

No packaged release in the registry yet. Follow the repository README.

RepositoryPublisher installStep-by-step: add this in Claude

Remote URL or a command on your machine? How the two installs differ

Run on PlugMCP

Host a copy with our keys and logs.

A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.

Launch

Server facts

Local

Transport: stdio
License: BSD-2-Clause
Last checked: 2026-10-28

Stars: 3
Tools: —
Uptime: —

Overview

About Protege

Tools

No tool list yet. Remote servers get one after the first health check.

FAQ

Questions

How do I install Protege?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Protege need an API key?

No required environment variable is listed on this record.

What transport does Protege use?

Protege runs locally over stdio.

When was Protege last checked?

From the publisher

Documentation

Protégé MCP

📖 Documentation

Guide	What's inside
Installation	Requirements (Java 17+), manual install, and Check for plugins.
Connecting a client	The server model (ports, OAuth vs. token) + Claude Code, Codex CLI, VS Code, Claude Desktop recipes.
Ontology Assistant	The in-Protégé chat: what it is, attachments, privacy, settings.
Tools	All 105 tools by category.
Prompts	The 11 guided workflows available to MCP clients.
Project policy & QC	Policy v1/v2/v3 discovery and validation, workspace membership, 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, and 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 · Changelog	The release scheme and notes for every version.

Highlights

105 structured tools + 11 guided prompts over the live, active ontology — explore, edit, curate, govern, extract modules, run SWRL rules, SPARQL and SHACL, and reason.
0.8.0 release capabilities — governed SSSOM mappings, external-term reuse proposals, exact reasoner capability reports, preview-first inference materialization, and bounded asynchronous jobs.
Enforced project boundaries (0.6.1) — direct paths are confined to the canonical project root with request-scoped filesystem capabilities, policy network/import controls block unapproved fetches, locked-import checksums are automatic gates, and verified rollback rejects a failing delta before it reaches the live ontology.
Least-privilege OAuth — all 105 tools declare and enforce ontology, release, filesystem, network, and server capabilities from the propagated principal. Explicit read grants cannot mutate, release, or open caller-selected files; existing mcp grants and the static token retain the documented local-admin compatibility profile.
Attributable local operations — every authorized, denied, successful, and failed tool call is written to a rotated owner-only stream without tokens, prompts, attachments, or ontology content; export_audit_log produces an explicitly confirmed, project-contained review artifact.
One endpoint, any number of windows (0.5.0) — a shared broker owns the configured port across every Protégé window and instance and routes each MCP session to a live window, so one fixed URL (http://127.0.0.1:8123/mcp) always works.
Safe, testable authoring (0.4.0, strengthened in 0.6.1) — apply_changes verify=report|rollback runs the isolated project/change-set gate before commit; a re-runnable competency-question suite and SPARQL invariants (verify_ontology) catch regressions; a single run_qc_suite gate rolls up reasoner + profile + structural + invariants + CQ checks; and search_entities ranks policy-configured RDFS/SKOS/OBO preferred labels and synonyms, explains each match/collision, and keeps fuzzy/synonym reuse review-only so an assistant grounds a term before minting.
Reproducible project QC (0.6.0) — discover and validate a checked-in policy, fingerprint the active ontology and loaded imports, resolve persisted invariant/CQ/multi-SHACL assets inside a confined project root, and run reasoner/profile/structural/governance/invariant/CQ/SHACL checks over one isolated snapshot with distinct pass / policy fail / execution error outcomes. The selected reasoner's exact Protégé configuration is preserved without classifying or mutating the live reasoner.
Portable project identity (0.6.0) — validate RO-Crate 1.0–1.3 metadata and report a W3C RDFC-1.0 + SHA-256 identity for the asserted root RDF dataset, separately from the editor revision fingerprint and imported-artifact checksums.
Import integrity inspection (0.6.0) — inspect deterministic direct/transitive import graphs, missing documents, cycles, source locations, and version conflicts; document loads keep the compatible warning default and support strict missing_imports=error for project/release workflows.
Edits are GUI-visible and undoable — every edit goes through Protégé's shared OWLModelManager, so it appears in the editor immediately and joins the Edit ▸ Undo stack.
Safe by default — a local server, bound to loopback (127.0.0.1) unless you change the bind-address preference, that requires OAuth or a bearer token on every request, with optional read-only mode and confirm-each-write.
In-app Ontology Assistant — chat that drives your existing claude, codex, agy, or opencode CLI back into the server with one short-lived, attributable credential per turn instead of the static admin token; no API key is stored by Protégé.

Requirements

Install

Download protege-mcp-<version>.jar from the latest release.
Drop it into Protégé's plugins/ folder (keep only one such jar — the bundle is a singleton).
Restart Protégé on a Java 17+ JVM.

Prefer automatic discovery and updates? Use Protégé's File ▸ Check for plugins. Full instructions, including per-OS plugin paths, are in the Installation guide.

Quick start

Settings ▸ MCP — configure and start the server (default http://127.0.0.1:8123/mcp). The MCP Server view shows the bound URL and bearer token.
Either connect an MCP client (Claude Code, Codex CLI, Antigravity, OpenCode, VS Code, Claude Desktop) or open the Ontology Assistant tab and type a request — "What classes are in this ontology?", then "Create a class FooBar under Thing."

Building from source

Requires Git, Apache Maven, and a JDK 17+:

git clone https://github.com/hakjuoh/protege-mcp.git
cd protege-mcp
mvn clean package

License

The software is licensed under the BSD 2-Clause License.