# AI Scanner [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/Aakashbhardwaj27/ai-scanner-mcp  
**GitHub Stars:** 1  
**npm Downloads (last month):** 182  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/ai-scanner

## Description
Scan codebases for LLM/AI SDK usage, exposed API tokens, and hardcoded secrets.

## Tools
Capabilities this server exposes over MCP:

- **scan_directory** — Scan a directory for LLM SDK usage, AI frameworks, exposed API tokens, and hardcoded secrets. Returns all findings grouped by type (token, secret, sdk, framework, endpoint, model) with file locations and severity levels.
- **check_secrets** — Security check — scan a directory for exposed API tokens and hardcoded secrets only. Returns pass/fail with a list of any exposed credentials found. Use this before committing code or in CI pipelines.
- **ai_inventory** — Get an inventory of all AI/LLM technologies used in a codebase — which SDKs are imported, which frameworks are integrated, which models are referenced, and which provider APIs are called. No secret detection, just awareness.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "ai-scanner": {
    "command": "npx",
    "args": ["-y","ai-scanner-mcp"]
  }
}
```

## Documentation

## What AI Scanner MCP server does

The AI Scanner MCP server lets an MCP-compatible AI client inspect a local codebase for AI usage and security findings. It exposes three tools with distinct purposes:

- `scan_directory` performs the broadest review, covering LLM SDKs, AI frameworks, exposed AI tokens, generic credentials, model references, and LLM endpoint URLs. Findings include severity levels.
- `check_secrets` performs a credential-focused pass/fail check. It is suited to checking a project before committing or pushing changes.
- `ai_inventory` describes the AI stack without looking for secrets, including detected SDKs, frameworks, models, and API endpoints.

The scanner advertises coverage for more than 20 AI token types, 59 generic secret patterns, 23 LLM SDKs, and 24 AI frameworks, with 145 detection patterns in total. Examples named by the project include OpenAI, Anthropic, Google, AWS, HuggingFace, Groq, Replicate, LangChain, LlamaIndex, CrewAI, AutoGen, DSPy, and Vercel AI SDK.

## How it works

An AI client sends a tool request containing the directory to inspect. The server passes that request to the underlying `ai-scanner` package and returns the relevant scan or inventory result to the client. Each tool requires a directory path.

`scan_directory` can be narrowed with `ai_only`, which skips generic credentials such as Stripe or GitHub tokens. Its `scan_env` option controls whether `.env` files are included. Endpoint and model detection are enabled by default and can be controlled with `include_endpoints` and `include_models`.

`check_secrets` also supports `ai_only` and `scan_env`. `ai_inventory` takes only the directory path and does not perform secret detection. The AI Scanner MCP server therefore supports either a security-oriented workflow or an AI-stack discovery workflow without combining their results.

## Setup and configuration

The package requires Node.js 18 or newer. Install or launch it through `npx` as an MCP server; the README provides configurations for Claude Code, Claude Desktop, Cursor, and Windsurf. The command is `npx ai-scanner-mcp`.

Claude Code uses `claude mcp add ai-scanner npx ai-scanner-mcp`. Claude Desktop, Cursor, and Windsurf use an MCP server entry whose command is `npx` and whose argument is `ai-scanner-mcp`. Claude Desktop stores its configuration under the platform-specific application-support location, while Cursor uses `.cursor/mcp.json` and Windsurf uses `~/.windsurf/mcp.json`.

## Tools and capabilities

Use `scan_directory` for a comprehensive review, `check_secrets` for a pass/fail credential gate, and `ai_inventory` when the goal is to understand how a project uses AI without secret scanning. The AI Scanner MCP server can inspect AI-specific tokens separately from generic secrets through the `ai_only` option.

## Limitations and notes

Results depend on the scanner's documented detection patterns. `ai_inventory` intentionally omits secret detection, so it is not a substitute for `scan_directory` or `check_secrets` when credential exposure is the concern. `.env` files are excluded unless `scan_env` is enabled. The provided setup documents local stdio-style launching with `npx`; no hosted service or remote deployment configuration is described.

_Full upstream README: https://allmcps.com/mcp/ai-scanner/readme_