SaglitzSecure
A security review your coding agent can actually run.
SaglitzSecure is a static security scanner that needs no parser. It reads web and backend source and Apple's Info.plist, and every finding it reports carries the file and line it came from plus a dated, primary source. Run it from the terminal, connect it to Claude or Cursor over MCP, or put it in front of a merge as a CI gate.
Why
Ask a model to review your code for security and it will find something — that is the problem. It produces confident, plausible advice at the same rate whether or not it knows anything about the topic, and nothing in the output distinguishes the two. SaglitzSecure stays quiet instead: if a subject is not in its knowledge base it says so and lists the nearest covered ones. No advice without a source.
Install
Node 20+. One command for a scan, a config block for an MCP client.
claude mcp add --scope user saglitzdesign -- npx -y saglitzdesign-mcp{
"mcpServers": {
"saglitzdesign": {
"command": "npx",
"args": ["-y", "saglitzdesign-mcp"]
}
}
}npx -y saglitzdesign-mcpEight tools
Scan, explain, plan the fix, then check the fix landed.
scan_projectRun the rule engine over a directory and return findings with file, line and a dated source for each one.
explain_findingWhy a specific finding matters, what it lets an attacker do, and what the primary source says.
security_guideThe guidance for one topic, in full — or, if the topic is not covered, the nearest ones that are.
knowledge_freshnessWhen each document was last verified against its source, so stale advice can be seen rather than trusted.
remediation_planAn ordered plan for the findings: what to fix first, and what each fix actually changes.
verify_fixesRe-run against the same project and report which findings are gone — the step most tools leave out.
threat_modelWalk the project's trust boundaries and name what crosses them.
deep_review_protocolThe manual protocol for the parts static rules cannot reach, as a checklist to work through.
The knowledge base
Eleven sourced documents across web, Apple platforms and doctrine, behind a 15-rule engine. Findings can be written as JSON, Markdown or SARIF, which is what lets a CI job fail a pull request on them.
Read the knowledge baseHeaders, transport, secrets in source, dependency and configuration risk.
Info.plist configuration, transport security, entitlement and permission surface.
How to weigh a finding, what evidence a claim needs, and when to stay quiet.
How you use it
Ask your agent to run a security analysis on the project.
“Using saglitzdesign, plan the design of an iOS fitness app.”
get_design_roadmap → a phased plan with the docs to read at each step.
“Review my landing page for conversion with saglitzdesign.”
design_review_checklist + get_design_examples.
“How should a primary button behave on mobile?”
get_component_guidance → specs, states, labels, anti-patterns.
“Measure this screenshot of my app.”
measure_screenshot → the real palette, exact contrast ratios, alignment and spacing, as numbers.
“Turn our Tailwind theme into SwiftUI tokens.”
import_design_tokens → the roles it names, the ones it is missing, and the set re-emitted for iOS.
“What's llms.txt and how do I set it up?”
seo_geo_guide (GEO) → the tactic with a ready-to-use example.
Frequently asked
How is this different from asking an AI to review my code?
A model answers every question at the same confidence, including the ones it should decline. This does the opposite: the rules are deterministic and the advice is bounded by a knowledge base, so a topic that is not covered produces silence and a list of what is, rather than something that reads correct and is not.
Does it need to parse my language?
No, and that is deliberate. It works over source as text plus configuration files, so a project in a language it has never seen still gets the rules that apply to it. The trade is that it finds patterns rather than proving properties — it is a reviewer, not a type system.
What can it reach on my machine?
Very little, by construction. The engine's only dependency is zod and its node builtins are allowlisted to crypto, fs/promises and path — so the part of the tool that reads your source cannot spawn a process or open a socket.
Can it fail a build?
Yes. SARIF output is the format CI systems already understand, so a job can gate a merge on the findings rather than printing them into a log nobody reads.
What does it cost?
Nothing. MIT-licensed, no account, no key.
A security review your coding agent can actually run.