Product · open source · MCP server

AI writes code.
AILC gets it to done.

Blueprints and a spec before the code. Oversight while you work. An honest verdict and fresh docs after. AILC connects with one line, picks the checks for your project itself, and does not say “done” until it is true. Everything runs on your machine; your code never leaves it.

How to connect View on GitHub

74 checks in one program
15 languages
100% offline

AILC · READY TO SHIP?~/project

Passwords and keysclean
Dangerous code (SAST)clean
Known library holes0
Stubs and TODOs2
Testsnot run

Quality score86/100NOT READY YET

Two stubs and tests never run. AILC will not write “done” until the check truly passes.

Why it earns trust

Never fakes “done”

The verdict is computed by strict rules. On the same code the answer is always identical, so you can show it to a manager or make it a condition for shipping.

Does not cry wolf

It refutes half the empty alarms itself: a key in a comment, a stub, a textbook example. Only real problems are flagged in red.

Sees what is hidden

It tracks the dangerous path of data across strings and functions, catching holes typical of code written by neural networks.

How it works

Three steps, all inside the editor you already have open.

01
Paste three lines
Into your editor config. The right binary for your system downloads itself. No build, no scripts, no Rust.
02
Ask in plain words
For example, “check whether everything is fine before shipping.” No tool names to learn: AILC picks the checks itself.
03
Get a clear verdict
A list of problems, a quality score, and a one-click fix for format and style. Dangerous is red, the rest are hints.

Catches holes typical of AI-written code

AILC follows where user input or a model’s reply ends up, and whether it reaches a dangerous place, even across several lines and renamings. Ordinary line-by-line checks miss this. It covers the OWASP list for AI apps: prompt injection and insecure handling of model output.

What AILC finds in your code

Not promises but specifics. 74 tools in nine families; here is the core.

Security

Hardcoded keys: AWS, GitHub, Stripe and 28 more kinds. OWASP Top 10 vulnerabilities. Dependency checks against known holes, offline via the OSV database.

Quality

Code smells, dead code, over-complex functions, cyclic dependencies, stubs and TODOs, undocumented code.

Docs and contract

Documentation stale against the code and broken public interfaces others depend on. Docs are regenerated straight from the code.

Russian compliance

Personal data in logs (152-FZ), storing data abroad (242-FZ), foreign trackers. These are signals for review, not a legal guarantee.

Deep analysis on 15 languages

PythonJavaScriptTypeScriptGoJavaRubyPHPC#RustKotlinScalaCC++SwiftDart

Connect in a minute

Already have Node.js? Paste the config. Nothing to build, no registration.

Claude Code, one command

claude mcp add ailc -- npx -y ailc-mcp

Cursor and Claude Code · .mcp.json

{
  "mcpServers": {
    "ailc": {
      "command": "npx",
      "args": ["-y", "ailc-mcp"]
    }
  }
}

VS Code · Copilot

{
  "servers": {
    "ailc": { "type": "stdio", "command": "npx", "args": ["-y", "ailc-mcp"] }
  }
}

From scratch? One command installs AILC and prints the config. Node and Docker not needed.

macOS · Linux

curl -fsSL https://raw.githubusercontent.com/pro-deploy/ailc/main/install.sh | sh

Windows

irm https://raw.githubusercontent.com/pro-deploy/ailc/main/install.ps1 | iex

On first connect AILC downloads itself and creates an .ailc folder in the project with rules and scaffolding. After that the AI agent calls the check itself whenever you ask it to look at the code.

Open and verifiable

For a developer, open code and honesty weigh more than pretty logos.

639tests, including a false-positive corpus and a 15-language matrix

Apache 2.0open core: fork it, extend it, build it yourself

Honesta public limitations section and a roadmap, no varnish

Honestly: AILC does not replace human review or a pentest, and legal signals are not a legal guarantee. Deep analysis works on 15 languages; for rare ones the checks are simpler. We say this plainly because trust is worth more than pretty promises.

Connect in a minute and write with confidence

One line in your config, the ready binary downloads itself. Then AILC runs the cycle: blueprints before the code, oversight during, an honest verdict and fresh docs after.