MCP001 · MCP002 · MCP005
Code execution
Shell input, executable selection, and dynamic evaluation.
Live now · CLI, GitHub Action, report explorerv0.6.3 · Ubuntu + Windows tested
See where agent-connected tools can reach—and what changed.
Uleravo turns MCP server code and local scan reports into redacted, reviewable security evidence. Scan locally, run it in GitHub Actions, or explore a JSON report privately in your browser. Skills, Plugins, and harness permission deltas are next—not current production coverage.
Apache-2.0 core. No account or API key. Node.js 22.13 or newer. Founding pilot details.
install / scannode ≥ 22.13
01 · Install
npm install -g ./uleravo-0.6.3.tgz02 · Scan
uleravo scan ./path-to-mcp-serversample resultscan complete
MCP001 · shell inputcriticalMCP003 · filesystem pathhighMCP007 · credentialhigh
12
focused checks
165
passing tests
25
internal validation cases
0
target code executed
Focused by design
Evidence you can act on.
Each finding includes a stable fingerprint, source location, redacted evidence, confidence, remediation, and defensible CWE, OWASP, or MITRE ATLAS mappings.
MCP003 · MCP004 · MCP008 · MCP009
Data reach
Filesystem paths, outbound destinations, and unsafe transport.
MCP007 · MCP010 · MCP011
Trust material
Credentials, mutable packages, and whole-environment forwarding.
MCP006 · MCP012
Tool behavior
Suspicious instructions and undeclared destructive behavior.
Static and local
The scanner does not import target code, run package scripts, contact discovered URLs, or invoke Git.
Reports before thresholds
JSON, SARIF, comparison, and signed-report workflows preserve evidence before an enforcement decision is made.
Incomplete means error
Unreadable inputs, safety limits, unresolved LFS pointers, and analysis failures cannot silently become a clean result.
Clear product boundary
What is live—and what is not.
Today's analyzer finds 12 focused risk patterns in MCP implementations. The browser explorer reads an existing Uleravo report; it does not rescan source. Skills and Plugins inventory plus Codex harness permissions are unreleased development work.
Uleravo does not yet observe runtime behavior or provide a security enforcement boundary. Static analysis is not certification, and report files should still be treated as sensitive even though the private explorer keeps them in your tab.
Start free, improve with users
A useful release now. A team product built from real demand.
Free CLI beta
Use it today on your own machine or CI runner.
$0
- All 12 current checks
- Text, JSON, and SARIF evidence
- Report comparison and offline signing
- Unlimited local scans
Founding Team Pilot
Hands-on rollout while measured use determines whether and how a hosted team layer should be built.
$500/ initial 30 days
- Guided rollout for up to 10 repositories
- Weekly critical and high finding triage
- Direct support and priority review of reproducible precision issues
- Product feedback may inform future analyzers; no feature or delivery date is included
Applications are open. Work begins only after a signed scope and data-handling agreement; there is no self-serve checkout.
Continuous releases
Launch is the starting line.
- Now
Evidence workbench
The public CLI and GitHub Action scan MCP code without executing it. The private browser explorer turns a local JSON report into a reviewable view without uploading the report.
- Next
Agent capability inventory
Snapshot Agent Skills and Plugins, inspect one defined Codex configuration harness, normalize declared and effective capabilities, and show meaningful permission changes between versions. This work is in development, not released coverage.
- Later
Runtime assurance and team control
Add opt-in runtime observation, shared policy, approvals, audit history, alerts, and integrations after the required tenant-isolation, retention, and incident-response controls are in place.
Release integrity
Verify the download.
SHA-256 for uleravo-0.6.3.tgz · 144,768 bytes
Download checksum filef99c30a2f38487693ddf35e72bf6ace01210538de1f3c534ff965e057e3081f0