Skip to content

Releases

v0.1.0 — Phase 1 MVP

Choose a tag to compare

@moondoom moondoom released this 27 Oct 13:22

AppSecGate MCP v0.1.0 — Phase 1 MVP

First public release of AppSecGate MCP: a local-first security scanner for AI coding agents. Runs inside Claude Code, Cursor, Cline, and Continue. Offline, no network calls, no LLM calls on the server.

What's in this release

  • 14 MCP tools: scan_file, scan_diff, scan_project, get_fix_context(_batch), verify_fix(_batch), query_findings_db, export_snapshot, get_config, update_config, submit_feedback, get_scan_history, suppress_finding.
  • Three bundled scanners: Semgrep (SAST), Gitleaks + TruffleHog (secrets), Trivy (SCA). All offline — Semgrep rules and Trivy DB are seeded at image build time.
  • Deterministic fingerprinting: findings are deduplicated across scans, so re-scanning the same file doesn't create duplicates.
  • Shown findings log: every finding the agent sees is recorded in shown_findings.db. This is the foundation for the Phase 2 CI gate (block_on_unresolved_local).
  • Fix context workflow: the agent gets code window + rule doc via get_fix_context, applies a fix with its own model, and confirms via verify_fix. The server never calls an LLM.

Docker image

Pull the image:

docker pull moondoom/appsecgate-mcp:0.1.0
docker pull moondoom/appsecgate-mcp:latest

Both tags point to the same digest. 0.1.0 is immutable; latest will move with future stable releases.

Image size: ~2.5 GB (bundled Semgrep rules + Trivy DB + Gitleaks + TruffleHog for fully offline operation).

What's NOT in this release (Phase 2 and later)

  • CI gating — Phase 1 is warn-only locally. Blocking in CI, including block_on_unresolved_local, is Phase 2.
  • Master server — no central sync, no team dashboards, no ASOC/DefectDojo integration yet.
  • RBAC — no roles, no approval workflows.
  • Web UI — no GUI yet.
  • Remote transport — stdio only. HTTP transport is Phase 2.

If you need any of the above, Phase 1 is not for you yet. If you want local, offline, agent-driven scanning with fix context — it is.

Known limitations

  • Large image — 2.5 GB. Bundled scanners trade size for offline operation. If you need a smaller image, build from source with only the scanners you need.
  • Rules are frozen at build time — no online rule updates. This is intentional (offline by design), but it means Semgrep rules and Trivy DB need a rebuild to update.
  • Cline + snap VS Code — path resolution for the project mount may require an absolute path in cwd and -v. See the deploy-agents guide for workarounds.
  • Three test runs per model in the benchmark — the model comparison in the article is a pilot, not a statistically significant benchmark.

Install

Full instructions for all four clients (Claude Code, Cursor, Cline, Continue):

https://github.com/moondoom/appsecgate-mcp/blob/main/docs/deploy-agents.md

Quick check: restart your client, then call get_config or scan_file on any file. Tools appear with the mcp__appsecgate__ prefix.

Verification

If you want to verify the image before running:

The image runs as non-root (UID 1000), with --read-only rootfs, --cap-drop=ALL, --network=none, and the project mounted :ro. It never writes to your source tree.

What's next

Phase 2: master server, CI integration, block_on_unresolved_local gate. Track progress in docs/open-tasks.md and the product vision doc:

Feedback, bug reports, and questions — open an issue.

License

MIT. Bundled scanner rules are third-party — see the third-party notices in the repository:

https://github.com/moondoom/appsecgate-mcp/blob/main/mcp-server/rules/THIRD_PARTY_NOTICES.md