PortunusMCP Sentinel v1.3.0

PortunusMCP Sentinel 1.3.0 adds explicit rules-only scanning and includes the accepted Phase 16–18 correctness and first-use changes.

pipx install portunusmcp-sentinel==1.3.0
sentinel scan . --rules-only

• Rules-only: Python and TypeScript deterministic scans bypass model clients, review cache, network calls, Docker, runtime configuration, and target execution. Select with --rules-only, SENTINEL_RULES_ONLY, or [scanner].rules_only; CLI overrides environment, which overrides project configuration. Existing scan and Action defaults remain unchanged.
• First use: scanning needs no initialization. Default sentinel init writes only permissions. Python runtime scaffolding moves to sentinel init --dynamic, followed by sentinel scan . --no-rules-only; upgrades preserve validated existing permissions. Dynamic scans need Docker and model review, which transmits bounded source context and incurs model cost. TypeScript remains static-only.
• Integrations: the Action adds an inheritable rules-only input and pins the 1.3.0 package. The existing pre-commit hook uses rules-only. Fork credential restrictions and SARIF upload policy are preserved. Installation, dependency auditing, and eligible Action uploads have separate network requirements.
• Static correctness: supported same-file helper flows and value-specific validation, authentication, integrity, and sanitizer checks prevent unrelated checks from exempting unsafe inputs.
• Runtime evidence: probes require valid baselines and observable security violations. Per-probe outcomes and confirmed effects survive review; incomplete analysis returns exit 3.

--allow-degraded remains a fallback policy and can still call the model; use --rules-only to disable review. Findings retain thresholds, inline suppression, and baselines. Completed scans exit 0 below the threshold or 1 for threshold findings; input errors exit 2 and incomplete analysis exits 3. Completion does not prove a target is secure.

Compatibility: native JSON is 1.5.0, with nullable finding/provenance reviews and a null GPT summary in rules-only reports. Consumers requiring review objects must update. SARIF remains 2.1.0 and now exports stage records; older SARIF without stages remains readable. Baseline-v2 retains supported migration from native 1.3/1.4 reports without inventing runtime proof. Stable rule IDs are preserved.

• Changelog
• Phase 18 verification
• PyPI
• Documentation

Release commit: c45e10cf878cfa0f926687f88a5679729ce9a19f.

Verification: all 44 release jobs passed, including TestPyPI promotion, PyPI hashes and attestations, Linux/macOS/Windows Python 3.10–3.13 public installs, installed-wheel Docker replay, and mandatory network-isolated rules-only scans. All 27 main CI jobs passed. No paid model calls were used.

The exact v1.3.0 Action proof and signed v1 alias proof passed with complete rules-only reports, null model review, zero findings, and validated SARIF upload. Both signed tags resolve to the release commit. The older demo control was corrected to consume its validated model value after the first proof exposed a low-severity candidate; the scanner package was unchanged. These checks establish release integration, not broad detection accuracy.

SHA-256:

e5c529967c58735303b3ab5d6a836607d85f5f078949587a1335c5d6544d6cbf  portunusmcp_sentinel-1.3.0-py3-none-any.whl
859022c18bfd62f19983f98a40ad7ddb6ee14b655462f7217d6defd4acfd9523  portunusmcp_sentinel-1.3.0.tar.gz

PortunusMCP Sentinel v1.2.1

PortunusMCP Sentinel 1.2.1

PortunusMCP Sentinel scans MCP server source before deployment and pairs static
candidates with GPT review, isolated Python runtime probes, and SARIF reports.

This maintenance release brings the PortunusMCP branding to installed console
output, adds the public documentation and pinned real-server walkthrough, and
fixes cold Semgrep startup exhausting the Windows scan deadline. Private
vulnerability reporting and contributor/maintenance guidance are now available.
No scanner capabilities, dependencies, stable rule IDs, or Action inputs/outputs
were added. Native JSON remains 1.4.0 and SARIF remains 2.1.0.

pipx install portunusmcp-sentinel==1.2.1
sentinel scan /path/to/server --static-only --allow-degraded

Rules-only Python/TypeScript scans need no key or Docker. Static GPT review needs
an operator key; full probing additionally needs Docker and a Python target.
sentinel demo --replay-review runs fresh isolated fixture probes with recorded
GPT responses, without new model calls.

• Walkthrough and coverage limits
• Changelog
• PyPI
• Marketplace
• Security policy
• 2:45 demo: older MCP Sentinel v0.1.0 branding and historical alerts

Verified release commit: 9fae385c684781f12702f50cbae60a6cfc48c867.

• Release workflow: all 43 jobs passed, including TestPyPI verification, protected PyPI promotion, provenance, and Linux/macOS/Windows Python 3.10–3.13 installs. Four initial public installs encountered stale PyPI index responses and passed on a failed-job-only retry; no artifacts were republished.
• Public artifact provenance: wheel and source archive.
• Exact v1.2.1 clean Action proof and signed v1 clean proof passed with complete, validated, uploaded SARIF and zero findings. Both signed refs resolve to the release commit.
• The published-package walkthrough reproduced zero findings across five files; SENT-001 was skipped for absent permissions. This is not a claim that the target is vulnerability-free.
• Current positive evidence comes from installed-wheel Docker replay, with recorded GPT responses and fresh probes. Historical positive code-scanning upload proof remains v0.1.0, verified 2026-07-21. The new clean Action runs are integration proof, not fresh positive-alert proof.

SHA-256:

143230824f214ac8d374d3c9850d2de8e4f568a069bc227f326722376140be94  portunusmcp_sentinel-1.2.1-py3-none-any.whl
753ea10b8256e0d91a9902d0473b817a73e740c5daca992360f6bb0a4fd5a38a  portunusmcp_sentinel-1.2.1.tar.gz

MCP Sentinel v1.2.0

MCP Sentinel v1.2.0 ships the Phase 10–12 release: configurable GPT review endpoints, TypeScript static analysis, and team adoption workflows.

• Adds incremental JSON baselines with exact static/dynamic identity matching.
• Adds reason-required Python and TypeScript inline suppressions.
• Adds baseline-aware console, JSON 1.4.0, SARIF 2.1.0, and GitHub Action output.
• Publishes the MCP Sentinel pre-commit hook.

Verification:

• Trusted PyPI publishing, provenance, and install matrix
• PyPI package
• Exact v1.2.0 adoption proof
• Signed v1 alias proof
• Usage and SHA-pinning guidance

MCP Sentinel v1.0.0

MCP Sentinel v1.0.0 is the first stable GitHub Marketplace release.

• Runs static analysis, required GPT semantic review, and sandboxed dynamic probing.
• Reports SENT-001 through SENT-011 with OWASP Agentic Top 10 mappings.
• Installs the exact portunusmcp-sentinel==1.0.0 package from PyPI.
• Supports uses: BashaarJavaid/MCP-Sentinel@v1.

Verification:

• Trusted PyPI publishing and provenance
• PyPI package
• Live clean/vulnerable Action proof
• Usage and SHA-pinning guidance

MCP Sentinel v0.1.0

MCP Sentinel v0.1.0 is the first public release of the build-time security scanner for Model Context Protocol servers.

Highlights

• Hybrid AST and Semgrep static analysis with SENT-001 through SENT-007.
• Required GPT-5.6 semantic review with live, replay, and explicitly degraded modes.
• Docker-isolated adversarial probing with SENT-008 through SENT-011.
• Canonical console, JSON, and offline-validated SARIF 2.1.0 reports.
• Composite GitHub Action with Code Scanning integration.
• Bundled fixtures, schemas, GPT replay cassettes, and judge-facing demo artifacts.

Quick start

Download the attached wheel, install it with pip or pipx, start Docker, and run:

sentinel demo --replay-review

Python 3.10 through 3.12 is supported. Replay mode does not require an OpenAI API key.

Release evidence

The attached wheel passed the full Linux, macOS, and Windows CI matrix, pip and pipx installation smoke tests, and the installed-wheel Docker replay test in GitHub Actions run 29686427335.

Wheel SHA-256: 4672e63413e87bf750113c06a21133162d00f1e71ca6259a8394028c22b677aa.