Visualize the risk. Understand the breach.
BreachFrame is an open-source desktop application for offline security auditing of access paths. You import configuration files: Kubernetes RBAC, AWS IAM, MCP agent configurations, or your own environment model. BreachFrame builds a permission graph, finds the paths that violate the policies you declare, flags risky permissions and MCP setups, and exports reports.
It is a defensive and educational tool. It does not scan networks, contact hosts or exploit anything. It only reads files you choose (or the bundled labs), and it never executes commands found in them.
Screenshots and the demo were captured from the real application on Windows 10 by the end-to-end test suite (
SCREENSHOTS=1 npm run test:e2e). Linux and macOS screenshots have not been captured yet.
| Module | What it does |
|---|---|
| Security Dashboard | Live totals from the local database: open findings by severity, risk score per analysis, workflow status, most frequent issues, projects, recent activity. |
| Attack Path Visualizer | Interactive graph of users, groups, roles, service accounts, services, APIs, AI agents, MCP servers, tools and protected resources. Lists every path that violates a deny policy, highlights it hop by hop, and supports filtering by node kind, search and "path only" mode. |
| Permission Auditor | Effective reach of every principal (what it can reach, how many sensitive resources, worst finding), plus RBAC, IAM and policy findings. |
| MCP Security Analyzer | Static analysis of MCP client configs (Claude Desktop, Claude Code, VS Code, Cursor, Windsurf, Gemini CLI). Covers hardcoded secrets (redacted), plaintext HTTP, unpinned packages, inline shells, curl | sh, privileged containers, broad filesystem roots, auto-approved tools and more. It can discover config files on your machine (you choose whether to import them). Nothing is executed or contacted. |
| Simulation Labs | Four self-contained training scenarios: MCP confused deputy, Kubernetes CI escalation, AWS iam:PassRole escalation, multi-tenant API BOLA. Each has a briefing, an attack graph, a vulnerable and a fixed variant, and can be imported as a project. |
| Blast-radius simulation | Pick a start node and replay, hop by hop, what an attacker controlling it could reach in the model. Always labelled model-based: no real system is touched. |
| Before & After | Compare two analyses, or a lab's vulnerable and fixed variants: risk delta, resolved, introduced and still-present findings, and graph edges removed or added. |
| Findings Manager | Every finding with severity, evidence, impact and recommendation. Workflow status (open, in progress, resolved, accepted risk, false positive) and analyst notes persist across re-analyses through stable fingerprints. History mode includes findings that are no longer present. |
| Report Generator | Self-contained HTML report (no scripts, print-friendly, light/dark) and machine-readable JSON, including artifact SHA-256 hashes, policies, methodology, finding statuses and notes. |
| CLI | breachframe analyze runs the same engine for CI pipelines, with --fail-on high to break builds. |
Other details:
- Project tabs, command palette (
Ctrl/⌘ K) and keyboard shortcuts. - Activity panel and loading states throughout.
- All data stays in a local SQLite database.
| Input | Detection |
|---|---|
| BreachFrame environment file (YAML/JSON) | breachframe: 1. Your own graph and deny-path policies, see docs/environment-format.md |
| Kubernetes RBAC manifests | Role, ClusterRole, RoleBinding, ClusterRoleBinding, ServiceAccount, List |
| AWS IAM | Policy documents and aws iam get-account-authorization-details output |
| MCP configuration | mcpServers / servers / mcp.servers / projects.*.mcpServers, JSON with comments |
All rules are listed in docs/rules.md. Policies are evaluated over the merged graph of all files in a project, so one policy can span environment, RBAC, IAM and MCP nodes.
Only what has actually been verified is marked as such.
| Platform | Status |
|---|---|
| Windows 10 x64 | ✅ Built, unit- and E2E-tested on Windows 10 Pro 22H2 (19045). NSIS and MSI installers built locally; silent per-user NSIS install, first launch and uninstall verified (the MSI was built but not install-tested). |
| Windows 11 x64 | |
| Linux x86_64: Ubuntu 22.04+ / Debian 12+ / Linux Mint 21+ | |
| Linux x86_64: Fedora / Arch | |
| Linux ARM64 | |
| macOS 10.15+ Apple Silicon / Intel |
Reports of success or failure on other platforms are very welcome. Please open an issue.
Installers for the latest version, plus SHA256SUMS.txt, are on the
Releases page:
| OS | File |
|---|---|
| Windows | BreachFrame_<version>_x64-setup.exe (NSIS, per-user) or BreachFrame_<version>_x64_en-US.msi |
| Linux | .AppImage, .deb, .rpm |
| macOS | BreachFrame_<version>_aarch64.dmg (Apple Silicon), _x64.dmg (Intel) |
Linux ARM64 packages (aarch64.AppImage, _arm64.deb, aarch64.rpm) are
attached too. You can also build from source (docs/building.md).
Windows:
- Run the setup
.exe(no administrator rights needed) or the.msi. - WebView2 is downloaded automatically if it is missing.
- Unsigned builds show a SmartScreen warning: More info → Run anyway.
- Uninstall from Settings → Apps.
Linux:
# Debian / Ubuntu / Mint
sudo apt install ./BreachFrame_<version>_amd64.deb
# Fedora
sudo dnf install ./BreachFrame-<version>-1.x86_64.rpm
# Any distribution (needs WebKitGTK 4.1 and GTK 3 installed)
chmod +x BreachFrame_<version>_amd64.AppImage && ./BreachFrame_<version>_amd64.AppImagemacOS:
- Open the
.dmgand drag BreachFrame to Applications. - Unsigned builds are blocked by Gatekeeper on first launch. Right-click the app
→ Open, or run
xattr -dr com.apple.quarantine /Applications/BreachFrame.app.
Verify a download with sha256sum -c SHA256SUMS.txt --ignore-missing.
Your data lives in a local SQLite database. See docs/architecture.md for its location.
- Try a lab: Simulation Labs → CI service account to cluster-admin. Open Attack graph and click a path, then press Simulate attacker. Switch to Fixed and then Before & After.
- Audit your own files:
- Projects → New project (
Ctrl+N). - Import files (
Ctrl+I) or paste text. On the MCP Analyzer page you can import MCP configs discovered on this machine. - Run analysis (
Ctrl+Enter).
- Projects → New project (
- Triage: open Findings, set a status, add notes. Statuses survive re-analysis, and fixed findings are resolved automatically.
- Report: Reports → Export HTML / JSON.
Ready-made inputs are in examples/. For example,
examples/environment/support-agent-platform.yaml
models an AI support agent whose MCP tool can reach a billing database:
policies:
- id: P-AGENT-01
name: Public-facing agents must not reach the billing database
severity: critical
deny:
from: { attributes: { exposure: public } }
to: { ids: ["db:billing"] }breachframe analyze examples/*/* # summary
breachframe analyze k8s/*.yaml --fail-on high # exit code 2 if high+ findings
breachframe analyze iam.json --format html --output report.html
breachframe lab k8s-ci-escalation --fixedThe CLI is built with cargo build -p breachframe-cli --release. It is not yet
shipped in the installers.
| Keys | Action |
|---|---|
Ctrl/⌘ K |
Command palette / search |
Ctrl/⌘ 1…9 |
Switch view |
Ctrl/⌘ N / I / Enter |
New project / import files / run analysis |
Ctrl/⌘ W, Ctrl/⌘ Shift ] [ |
Close / cycle project tabs |
Ctrl/⌘ . |
Activity panel |
? |
Shortcut help |
React + TypeScript UI ──Tauri IPC──► Rust app layer (commands, services, SQLite)
│
▼
breachframe-engine (pure Rust: parse → graph → rules →
policies → score → simulate/compare → HTML/JSON report)
▲
breachframe CLI
- Stack: Tauri 2, Rust, React 19, TypeScript, Tailwind CSS 4, React Flow, SQLite (bundled), Vitest, Playwright.
- No Python and no runtime downloads.
- The engine has no filesystem, network or process access.
See docs/architecture.md for details.
| Suite | Command | Coverage |
|---|---|---|
| Engine unit + integration | cargo test --workspace |
parsers, input limits, every rule, policy BFS, secrets, risk score, simulation, comparison, reports, labs, examples, app services (SQLite lifecycle) |
| Front end | npm test |
graph helpers, IPC contract (mocked Tauri IPC), shortcuts, UI components |
| Desktop E2E | npm run test:e2e |
the real Windows app over WebView2 DevTools: labs, simulation, project import, MCP view, findings workflow, reports, palette |
- SECURITY.md explains how to report vulnerabilities privately.
- The app enforces a strict Content Security Policy.
- Report previews run in a sandboxed iframe without scripts.
- Inputs are size-, depth- and type-limited.
- YAML alias bombs are rejected.
- Credentials are redacted before storage.
See CONTRIBUTING.md. Releases follow docs/releasing.md. They are draft-only and require maintainer approval.
Not implemented yet:
- Validate and publish signed installers for Windows and macOS
- Linux/macOS screenshots and validation on each listed distribution
- Ship the CLI inside the installers
- More importers: Azure RBAC, GCP IAM, Terraform plans, Kubernetes
Pod/ workload specs, OpenAPI security schemes - SARIF export for code-scanning integrations
- Graph export (PNG/SVG) from the visualizer
- Custom rule packs
- UI localisation (the app UI is English-only today)
MIT © BreachFrame contributors

