agentaudit 3.9.8 โ 3.9.10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +34 -34
- package/README.md +504 -504
- package/cli.mjs +1737 -1680
- package/index.mjs +605 -605
- package/package.json +45 -45
- package/prompts/audit-prompt.md +663 -663
package/README.md
CHANGED
|
@@ -1,504 +1,504 @@
|
|
|
1
|
-
<div align="center">
|
|
2
|
-
|
|
3
|
-
# ๐ก๏ธ AgentAudit
|
|
4
|
-
|
|
5
|
-
**Security scanner for AI packages โ MCP server + CLI**
|
|
6
|
-
|
|
7
|
-
Scan MCP servers, AI skills, and packages for vulnerabilities, prompt injection,
|
|
8
|
-
and supply chain attacks. Powered by regex static analysis and deep LLM audits.
|
|
9
|
-
|
|
10
|
-
[](https://www.npmjs.com/package/agentaudit)
|
|
11
|
-
[](https://agentaudit.dev)
|
|
12
|
-
[](LICENSE)
|
|
13
|
-
|
|
14
|
-
</div>
|
|
15
|
-
|
|
16
|
-
---
|
|
17
|
-
|
|
18
|
-
## ๐ Table of Contents
|
|
19
|
-
|
|
20
|
-
- [What is AgentAudit?](#what-is-agentaudit)
|
|
21
|
-
- [Quick Start](#-quick-start)
|
|
22
|
-
- [Commands Reference](#-commands-reference)
|
|
23
|
-
- [Quick Scan vs Deep Audit](#-quick-scan-vs-deep-audit)
|
|
24
|
-
- [MCP Server](#-mcp-server)
|
|
25
|
-
- [What It Detects](#-what-it-detects)
|
|
26
|
-
- [How the 3-Pass Audit Works](#-how-the-3-pass-audit-works)
|
|
27
|
-
- [CI/CD Integration](#-cicd-integration)
|
|
28
|
-
- [Configuration](#-configuration)
|
|
29
|
-
- [Requirements](#-requirements)
|
|
30
|
-
- [FAQ](#-faq)
|
|
31
|
-
- [Related Links](#-related-links)
|
|
32
|
-
- [License](#-license)
|
|
33
|
-
|
|
34
|
-
---
|
|
35
|
-
|
|
36
|
-
## What is AgentAudit?
|
|
37
|
-
|
|
38
|
-
AgentAudit is a security scanner purpose-built for the AI package ecosystem. It works in two modes:
|
|
39
|
-
|
|
40
|
-
1. **CLI tool** โ Run `agentaudit` in your terminal to discover and scan MCP servers installed in your AI editors
|
|
41
|
-
2. **MCP server** โ Add to Claude Desktop, Cursor, or Windsurf so your AI agent can audit packages on your behalf
|
|
42
|
-
|
|
43
|
-
It checks packages against the [AgentAudit Trust Registry](https://agentaudit.dev) โ a shared, community-driven database of security findings โ and can perform local scans ranging from fast regex analysis to deep LLM-powered 3-pass audits.
|
|
44
|
-
|
|
45
|
-
---
|
|
46
|
-
|
|
47
|
-
## ๐ Quick Start
|
|
48
|
-
|
|
49
|
-
<p align="center">
|
|
50
|
-
<img src="docs/cli-screenshot.png" alt="AgentAudit CLI โ discover and scan" width="700">
|
|
51
|
-
</p>
|
|
52
|
-
|
|
53
|
-
### Option A: CLI (recommended)
|
|
54
|
-
|
|
55
|
-
```bash
|
|
56
|
-
# Install globally (or use npx agentaudit)
|
|
57
|
-
npm install -g agentaudit
|
|
58
|
-
|
|
59
|
-
# Discover MCP servers configured in your AI editors
|
|
60
|
-
agentaudit
|
|
61
|
-
|
|
62
|
-
# Quick scan โ clones repo, checks code with regex patterns (~2s)
|
|
63
|
-
agentaudit scan https://github.com/owner/repo
|
|
64
|
-
|
|
65
|
-
# Deep audit โ clones repo, sends code to LLM for 3-pass analysis (~30s)
|
|
66
|
-
agentaudit audit https://github.com/owner/repo
|
|
67
|
-
|
|
68
|
-
# Registry lookup โ check if a package has been audited before (no cloning)
|
|
69
|
-
agentaudit lookup fastmcp
|
|
70
|
-
```
|
|
71
|
-
|
|
72
|
-
**Example output:**
|
|
73
|
-
```
|
|
74
|
-
AgentAudit v3.9.
|
|
75
|
-
Security scanner for AI packages
|
|
76
|
-
|
|
77
|
-
Discovering MCP servers in your AI editors...
|
|
78
|
-
|
|
79
|
-
โข Scanning Cursor ~/.cursor/mcp.json found 3 servers
|
|
80
|
-
|
|
81
|
-
โโโ tool supabase-mcp โ ok
|
|
82
|
-
โ SAFE Risk 0 https://agentaudit.dev/skills/supabase-mcp
|
|
83
|
-
โโโ tool browser-tools-mcp โ ok
|
|
84
|
-
โ โ not audited Run: agentaudit audit https://github.com/nichochar/browser-tools-mcp
|
|
85
|
-
โโโ tool filesystem โ ok
|
|
86
|
-
โ SAFE Risk 0 https://agentaudit.dev/skills/filesystem
|
|
87
|
-
|
|
88
|
-
Looking for general package scanning? Try `pip audit` or `npm audit`.
|
|
89
|
-
```
|
|
90
|
-
|
|
91
|
-
### Option B: MCP Server in your AI editor
|
|
92
|
-
|
|
93
|
-
Add AgentAudit as an MCP server โ your AI agent can then discover, scan, and audit packages using its own LLM. **No extra API key needed.**
|
|
94
|
-
|
|
95
|
-
<details>
|
|
96
|
-
<summary><strong>Claude Desktop</strong> โ <code>~/.claude/mcp.json</code></summary>
|
|
97
|
-
|
|
98
|
-
```json
|
|
99
|
-
{
|
|
100
|
-
"mcpServers": {
|
|
101
|
-
"agentaudit": {
|
|
102
|
-
"command": "npx",
|
|
103
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
104
|
-
}
|
|
105
|
-
}
|
|
106
|
-
}
|
|
107
|
-
```
|
|
108
|
-
</details>
|
|
109
|
-
|
|
110
|
-
<details>
|
|
111
|
-
<summary><strong>Cursor</strong> โ <code>.cursor/mcp.json</code> (project) or <code>~/.cursor/mcp.json</code> (global)</summary>
|
|
112
|
-
|
|
113
|
-
```json
|
|
114
|
-
{
|
|
115
|
-
"mcpServers": {
|
|
116
|
-
"agentaudit": {
|
|
117
|
-
"command": "npx",
|
|
118
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
119
|
-
}
|
|
120
|
-
}
|
|
121
|
-
}
|
|
122
|
-
```
|
|
123
|
-
</details>
|
|
124
|
-
|
|
125
|
-
<details>
|
|
126
|
-
<summary><strong>Windsurf</strong> โ <code>~/.codeium/windsurf/mcp_config.json</code></summary>
|
|
127
|
-
|
|
128
|
-
```json
|
|
129
|
-
{
|
|
130
|
-
"mcpServers": {
|
|
131
|
-
"agentaudit": {
|
|
132
|
-
"command": "npx",
|
|
133
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
134
|
-
}
|
|
135
|
-
}
|
|
136
|
-
}
|
|
137
|
-
```
|
|
138
|
-
</details>
|
|
139
|
-
|
|
140
|
-
<details>
|
|
141
|
-
<summary><strong>VS Code</strong> โ <code>.vscode/mcp.json</code></summary>
|
|
142
|
-
|
|
143
|
-
```json
|
|
144
|
-
{
|
|
145
|
-
"servers": {
|
|
146
|
-
"agentaudit": {
|
|
147
|
-
"command": "npx",
|
|
148
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
149
|
-
}
|
|
150
|
-
}
|
|
151
|
-
}
|
|
152
|
-
```
|
|
153
|
-
</details>
|
|
154
|
-
|
|
155
|
-
<details>
|
|
156
|
-
<summary><strong>Continue.dev</strong> โ <code>~/.continue/config.json</code></summary>
|
|
157
|
-
|
|
158
|
-
Add to the `mcpServers` section of your existing config:
|
|
159
|
-
```json
|
|
160
|
-
{
|
|
161
|
-
"mcpServers": [
|
|
162
|
-
{
|
|
163
|
-
"name": "agentaudit",
|
|
164
|
-
"command": "npx",
|
|
165
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
166
|
-
}
|
|
167
|
-
]
|
|
168
|
-
}
|
|
169
|
-
```
|
|
170
|
-
</details>
|
|
171
|
-
|
|
172
|
-
<details>
|
|
173
|
-
<summary><strong>Zed</strong> โ <code>~/.config/zed/settings.json</code></summary>
|
|
174
|
-
|
|
175
|
-
```json
|
|
176
|
-
{
|
|
177
|
-
"context_servers": {
|
|
178
|
-
"agentaudit": {
|
|
179
|
-
"command": {
|
|
180
|
-
"path": "npx",
|
|
181
|
-
"args": ["-y", "agentaudit", "--stdio"]
|
|
182
|
-
}
|
|
183
|
-
}
|
|
184
|
-
}
|
|
185
|
-
}
|
|
186
|
-
```
|
|
187
|
-
</details>
|
|
188
|
-
|
|
189
|
-
Then ask your agent: *"Check which MCP servers I have installed and audit any unaudited ones."*
|
|
190
|
-
|
|
191
|
-
---
|
|
192
|
-
|
|
193
|
-
## ๐ Commands Reference
|
|
194
|
-
|
|
195
|
-
| Command | Description | Example |
|
|
196
|
-
|---------|-------------|---------|
|
|
197
|
-
| `agentaudit` | Discover MCP servers (default, same as `discover`) | `agentaudit` |
|
|
198
|
-
| `agentaudit discover` | Find MCP servers in Cursor, Claude, VS Code, Windsurf | `agentaudit discover` |
|
|
199
|
-
| `agentaudit discover --quick` | Discover + auto-scan all servers | `agentaudit discover --quick` |
|
|
200
|
-
| `agentaudit discover --deep` | Discover + interactively select servers to deep-audit | `agentaudit discover --deep` |
|
|
201
|
-
| `agentaudit scan <url>` | Quick regex-based static scan (~2s) | `agentaudit scan https://github.com/owner/repo` |
|
|
202
|
-
| `agentaudit scan <url> --deep` | Deep audit (same as `audit`) | `agentaudit scan https://github.com/owner/repo --deep` |
|
|
203
|
-
| `agentaudit audit <url>` | Deep LLM-powered 3-pass audit (~30s) | `agentaudit audit https://github.com/owner/repo` |
|
|
204
|
-
| `agentaudit lookup <name>` | Look up package in trust registry | `agentaudit lookup fastmcp` |
|
|
205
|
-
| `agentaudit setup` | Register agent + configure API key | `agentaudit setup` |
|
|
206
|
-
|
|
207
|
-
### Global Flags
|
|
208
|
-
|
|
209
|
-
| Flag | Description |
|
|
210
|
-
|------|-------------|
|
|
211
|
-
| `--json` | Output machine-readable JSON to stdout |
|
|
212
|
-
| `--quiet` / `-q` | Suppress banner and decorative output (show findings only) |
|
|
213
|
-
| `--no-color` | Disable ANSI colors (also respects `NO_COLOR` env var) |
|
|
214
|
-
| `--help` / `-h` | Show help text |
|
|
215
|
-
| `-v` / `--version` | Show version |
|
|
216
|
-
|
|
217
|
-
### Exit Codes
|
|
218
|
-
|
|
219
|
-
| Code | Meaning |
|
|
220
|
-
|------|---------|
|
|
221
|
-
| `0` | Clean โ no findings detected, or successful lookup |
|
|
222
|
-
| `1` | Findings detected |
|
|
223
|
-
| `2` | Error (clone failed, network error, invalid args) |
|
|
224
|
-
|
|
225
|
-
---
|
|
226
|
-
|
|
227
|
-
## โ๏ธ Quick Scan vs Deep Audit
|
|
228
|
-
|
|
229
|
-
| | Quick Scan (`scan`) | Deep Audit (`audit`) |
|
|
230
|
-
|---|---------------------|---------------------|
|
|
231
|
-
| **Speed** | ~2 seconds | ~30 seconds |
|
|
232
|
-
| **Method** | Regex pattern matching | LLM-powered 3-pass analysis |
|
|
233
|
-
| **API key needed** | No | Yes (`ANTHROPIC_API_KEY` or `OPENAI_API_KEY`) |
|
|
234
|
-
| **False positives** | Higher (regex limitations) | Very low (context-aware) |
|
|
235
|
-
| **Detects** | Common patterns (injection, secrets, eval) | Complex attack chains, AI-specific threats, obfuscation |
|
|
236
|
-
| **Best for** | Quick triage, CI pipelines | Critical packages, pre-production review |
|
|
237
|
-
|
|
238
|
-
**Tip:** Use `agentaudit scan <url> --deep` to run a deep audit via the scan command.
|
|
239
|
-
|
|
240
|
-
---
|
|
241
|
-
|
|
242
|
-
## ๐ MCP Server
|
|
243
|
-
|
|
244
|
-
When running as an MCP server, AgentAudit exposes the following tools to your AI agent:
|
|
245
|
-
|
|
246
|
-
| Tool | Description |
|
|
247
|
-
|------|-------------|
|
|
248
|
-
| `audit_package` | Deep LLM-powered audit of a repository |
|
|
249
|
-
| `check_registry` | Look up a package in the trust registry |
|
|
250
|
-
| `submit_report` | Upload audit findings to the registry |
|
|
251
|
-
| `discover_servers` | Find MCP servers in local editor configs |
|
|
252
|
-
|
|
253
|
-
### Workflow
|
|
254
|
-
|
|
255
|
-
```
|
|
256
|
-
User asks agent to install a package
|
|
257
|
-
โ
|
|
258
|
-
โผ
|
|
259
|
-
Agent calls check_registry(package_name)
|
|
260
|
-
โ
|
|
261
|
-
โโโโโโดโโโโโ
|
|
262
|
-
โ โ
|
|
263
|
-
Found Not Found
|
|
264
|
-
โ โ
|
|
265
|
-
โผ โผ
|
|
266
|
-
Return Agent calls audit_package(repo_url)
|
|
267
|
-
score โ
|
|
268
|
-
โผ
|
|
269
|
-
LLM analyzes code (3-pass)
|
|
270
|
-
โ
|
|
271
|
-
โผ
|
|
272
|
-
Agent calls submit_report(findings)
|
|
273
|
-
โ
|
|
274
|
-
โผ
|
|
275
|
-
Return findings + risk score
|
|
276
|
-
```
|
|
277
|
-
|
|
278
|
-
---
|
|
279
|
-
|
|
280
|
-
## ๐ฏ What It Detects
|
|
281
|
-
|
|
282
|
-
<table>
|
|
283
|
-
<tr>
|
|
284
|
-
<td>
|
|
285
|
-
|
|
286
|
-
**Core Security**
|
|
287
|
-
|
|
288
|
-
|
|
289
|
-
|
|
290
|
-
|
|
291
|
-
|
|
292
|
-
|
|
293
|
-
|
|
294
|
-
|
|
295
|
-
</td>
|
|
296
|
-
<td>
|
|
297
|
-
|
|
298
|
-
**AI-Specific**
|
|
299
|
-
|
|
300
|
-
|
|
301
|
-
|
|
302
|
-
|
|
303
|
-
|
|
304
|
-
|
|
305
|
-
|
|
306
|
-
|
|
307
|
-
</td>
|
|
308
|
-
</tr>
|
|
309
|
-
<tr>
|
|
310
|
-
<td>
|
|
311
|
-
|
|
312
|
-
**MCP-Specific**
|
|
313
|
-
|
|
314
|
-
|
|
315
|
-
|
|
316
|
-
|
|
317
|
-
|
|
318
|
-
|
|
319
|
-
|
|
320
|
-
</td>
|
|
321
|
-
<td>
|
|
322
|
-
|
|
323
|
-
**Persistence & Obfuscation**
|
|
324
|
-
|
|
325
|
-
|
|
326
|
-
|
|
327
|
-
|
|
328
|
-
|
|
329
|
-
|
|
330
|
-
|
|
331
|
-
|
|
332
|
-
</td>
|
|
333
|
-
</tr>
|
|
334
|
-
</table>
|
|
335
|
-
|
|
336
|
-
---
|
|
337
|
-
|
|
338
|
-
## ๐ง How the 3-Pass Audit Works
|
|
339
|
-
|
|
340
|
-
The deep audit (`agentaudit audit`) uses a structured 3-phase LLM analysis โ not a single-shot prompt, but a rigorous multi-pass process:
|
|
341
|
-
|
|
342
|
-
| Phase | Name | What Happens |
|
|
343
|
-
|-------|------|-------------|
|
|
344
|
-
| **1** | ๐ **UNDERSTAND** | Read all files and build a **Package Profile**: purpose, category, expected behaviors, trust boundaries. No scanning yet โ the goal is to understand what the package *should* do before looking for what it *shouldn't*. |
|
|
345
|
-
| **2** | ๐ฏ **DETECT** | Evidence collection against **50+ detection patterns** across 8 categories (AI-specific, MCP, persistence, obfuscation, cross-file correlation). Only facts are recorded โ no severity judgments yet. |
|
|
346
|
-
| **3** | โ๏ธ **CLASSIFY** | Every finding goes through a **Mandatory Self-Check** (5 questions), **Exploitability Assessment**, and **Confidence Gating**. HIGH/CRITICAL findings must survive a **Devil's Advocate** challenge and include a full **Reasoning Chain**. |
|
|
347
|
-
|
|
348
|
-
**Why 3 passes?** Single-pass analysis is the #1 cause of false positives. By separating understanding โ detection โ classification:
|
|
349
|
-
|
|
350
|
-
- Phase 1 prevents flagging core functionality as suspicious (e.g., SQL execution in a database tool)
|
|
351
|
-
- Phase 2 ensures evidence is collected without severity bias
|
|
352
|
-
- Phase 3 catches false positives before they reach the report
|
|
353
|
-
|
|
354
|
-
This architecture achieved **0% false positives** on our 11-package test set, down from 42% in v2.
|
|
355
|
-
|
|
356
|
-
---
|
|
357
|
-
|
|
358
|
-
## ๐ CI/CD Integration
|
|
359
|
-
|
|
360
|
-
AgentAudit is designed for CI pipelines with proper exit codes and JSON output:
|
|
361
|
-
|
|
362
|
-
```yaml
|
|
363
|
-
# GitHub Actions example
|
|
364
|
-
- name: Scan MCP servers
|
|
365
|
-
run: |
|
|
366
|
-
npx agentaudit scan https://github.com/org/mcp-server --json --quiet > results.json
|
|
367
|
-
# Exit code 1 = findings detected โ fail the build
|
|
368
|
-
```
|
|
369
|
-
|
|
370
|
-
```bash
|
|
371
|
-
# Shell scripting
|
|
372
|
-
agentaudit scan https://github.com/owner/repo --json --quiet 2>/dev/null
|
|
373
|
-
if [ $? -eq 1 ]; then
|
|
374
|
-
echo "Security findings detected!"
|
|
375
|
-
exit 1
|
|
376
|
-
fi
|
|
377
|
-
```
|
|
378
|
-
|
|
379
|
-
### JSON Output Examples
|
|
380
|
-
|
|
381
|
-
```bash
|
|
382
|
-
# Scan with JSON output
|
|
383
|
-
agentaudit scan https://github.com/owner/repo --json
|
|
384
|
-
```
|
|
385
|
-
|
|
386
|
-
```json
|
|
387
|
-
{
|
|
388
|
-
"slug": "repo",
|
|
389
|
-
"url": "https://github.com/owner/repo",
|
|
390
|
-
"findings": [
|
|
391
|
-
{
|
|
392
|
-
"severity": "high",
|
|
393
|
-
"title": "Command injection risk",
|
|
394
|
-
"file": "src/handler.js",
|
|
395
|
-
"line": 42,
|
|
396
|
-
"snippet": "exec(`git ${userInput}`)"
|
|
397
|
-
}
|
|
398
|
-
],
|
|
399
|
-
"fileCount": 15,
|
|
400
|
-
"duration": "1.8s"
|
|
401
|
-
}
|
|
402
|
-
```
|
|
403
|
-
|
|
404
|
-
```bash
|
|
405
|
-
# Registry lookup with JSON
|
|
406
|
-
agentaudit lookup fastmcp --json
|
|
407
|
-
```
|
|
408
|
-
|
|
409
|
-
> **Coming soon:** `--fail-on <severity>` flag to set minimum severity threshold for non-zero exit (e.g., `--fail-on high` ignores low/medium findings).
|
|
410
|
-
|
|
411
|
-
---
|
|
412
|
-
|
|
413
|
-
## โ๏ธ Configuration
|
|
414
|
-
|
|
415
|
-
### Credentials
|
|
416
|
-
|
|
417
|
-
AgentAudit stores credentials in `~/.config/agentaudit/credentials.json` (or `$XDG_CONFIG_HOME/agentaudit/credentials.json`).
|
|
418
|
-
|
|
419
|
-
Run `agentaudit setup` to configure interactively, or set via environment:
|
|
420
|
-
|
|
421
|
-
```bash
|
|
422
|
-
export AGENTAUDIT_API_KEY=asf_your_key_here
|
|
423
|
-
```
|
|
424
|
-
|
|
425
|
-
### Environment Variables
|
|
426
|
-
|
|
427
|
-
| Variable | Description |
|
|
428
|
-
|----------|-------------|
|
|
429
|
-
| `AGENTAUDIT_API_KEY` | API key for registry access |
|
|
430
|
-
| `ANTHROPIC_API_KEY` | Anthropic API key for deep audits (Claude) |
|
|
431
|
-
| `OPENAI_API_KEY` | OpenAI API key for deep audits (GPT-4o) |
|
|
432
|
-
| `NO_COLOR` | Disable ANSI colors ([no-color.org](https://no-color.org)) |
|
|
433
|
-
|
|
434
|
-
---
|
|
435
|
-
|
|
436
|
-
## ๐ฆ Requirements
|
|
437
|
-
|
|
438
|
-
- **Node.js** โฅ 18.0.0
|
|
439
|
-
- **Git** (for cloning repositories during scan/audit)
|
|
440
|
-
|
|
441
|
-
---
|
|
442
|
-
|
|
443
|
-
## โ FAQ
|
|
444
|
-
|
|
445
|
-
### How do I set up AgentAudit?
|
|
446
|
-
|
|
447
|
-
```bash
|
|
448
|
-
npm install -g agentaudit
|
|
449
|
-
agentaudit setup
|
|
450
|
-
```
|
|
451
|
-
|
|
452
|
-
Or use without installing: `npx agentaudit`
|
|
453
|
-
|
|
454
|
-
### Do I need an API key?
|
|
455
|
-
|
|
456
|
-
- **Quick scan** (`scan`): No API key needed โ runs locally with regex
|
|
457
|
-
- **Deep audit** (`audit`): Needs `ANTHROPIC_API_KEY` or `OPENAI_API_KEY`
|
|
458
|
-
- **Registry lookup** (`lookup`): No key needed for reading; key needed for uploading reports
|
|
459
|
-
- **MCP server**: No extra key needed โ uses the host editor's LLM
|
|
460
|
-
|
|
461
|
-
### What data is sent externally?
|
|
462
|
-
|
|
463
|
-
- **Registry lookups**: Package name/slug is sent to `agentaudit.dev` to check for existing audits
|
|
464
|
-
- **Report uploads**: Audit findings are uploaded to the public registry (requires API key)
|
|
465
|
-
- **Deep audits**: Source code is sent to Anthropic or OpenAI for LLM analysis
|
|
466
|
-
- **Quick scans**: Everything stays local โ no data leaves your machine
|
|
467
|
-
|
|
468
|
-
### Can I use it offline?
|
|
469
|
-
|
|
470
|
-
Quick scans (`agentaudit scan`) work fully offline after cloning. Registry lookups and deep audits require network access.
|
|
471
|
-
|
|
472
|
-
### Can I use it as an MCP server without the CLI?
|
|
473
|
-
|
|
474
|
-
Yes! `npx agentaudit` starts the MCP server when invoked by an editor. The CLI and MCP server are the same package โ behavior is determined by how it's called.
|
|
475
|
-
|
|
476
|
-
### How does `discover` know which editors I use?
|
|
477
|
-
|
|
478
|
-
It checks standard config file locations for Claude Desktop, Cursor, VS Code, and Windsurf. It also checks the current working directory for project-level `.cursor/mcp.json` and `.vscode/mcp.json`.
|
|
479
|
-
|
|
480
|
-
---
|
|
481
|
-
|
|
482
|
-
## ๐ Related Links
|
|
483
|
-
|
|
484
|
-
- **Trust Registry**: [agentaudit.dev](https://agentaudit.dev)
|
|
485
|
-
- **Leaderboard**: [agentaudit.dev/leaderboard](https://agentaudit.dev/leaderboard)
|
|
486
|
-
- **Agent Skill**: [github.com/starbuck100/agentaudit-skill](https://github.com/starbuck100/agentaudit-skill) โ Full agent skill with pre-install security gate, detection patterns & peer review system
|
|
487
|
-
- **MCP Server Repository**: [github.com/starbuck100/agentaudit-mcp](https://github.com/starbuck100/agentaudit-mcp)
|
|
488
|
-
- **Report Issues**: [GitHub Issues](https://github.com/starbuck100/agentaudit-mcp/issues)
|
|
489
|
-
|
|
490
|
-
---
|
|
491
|
-
|
|
492
|
-
## ๐ License
|
|
493
|
-
|
|
494
|
-
[AGPL-3.0](LICENSE) โ Free for open source use. Commercial license available for proprietary integrations.
|
|
495
|
-
|
|
496
|
-
---
|
|
497
|
-
|
|
498
|
-
<div align="center">
|
|
499
|
-
|
|
500
|
-
**Protect your AI stack. Scan before you trust.**
|
|
501
|
-
|
|
502
|
-
[Trust Registry](https://agentaudit.dev) ยท [Leaderboard](https://agentaudit.dev/leaderboard) ยท [Report Issues](https://github.com/starbuck100/agentaudit-mcp/issues)
|
|
503
|
-
|
|
504
|
-
</div>
|
|
1
|
+
<div align="center">
|
|
2
|
+
|
|
3
|
+
# ๐ก๏ธ AgentAudit
|
|
4
|
+
|
|
5
|
+
**Security scanner for AI packages โ MCP server + CLI**
|
|
6
|
+
|
|
7
|
+
Scan MCP servers, AI skills, and packages for vulnerabilities, prompt injection,
|
|
8
|
+
and supply chain attacks. Powered by regex static analysis and deep LLM audits.
|
|
9
|
+
|
|
10
|
+
[](https://www.npmjs.com/package/agentaudit)
|
|
11
|
+
[](https://agentaudit.dev)
|
|
12
|
+
[](LICENSE)
|
|
13
|
+
|
|
14
|
+
</div>
|
|
15
|
+
|
|
16
|
+
---
|
|
17
|
+
|
|
18
|
+
## ๐ Table of Contents
|
|
19
|
+
|
|
20
|
+
- [What is AgentAudit?](#what-is-agentaudit)
|
|
21
|
+
- [Quick Start](#-quick-start)
|
|
22
|
+
- [Commands Reference](#-commands-reference)
|
|
23
|
+
- [Quick Scan vs Deep Audit](#-quick-scan-vs-deep-audit)
|
|
24
|
+
- [MCP Server](#-mcp-server)
|
|
25
|
+
- [What It Detects](#-what-it-detects)
|
|
26
|
+
- [How the 3-Pass Audit Works](#-how-the-3-pass-audit-works)
|
|
27
|
+
- [CI/CD Integration](#-cicd-integration)
|
|
28
|
+
- [Configuration](#-configuration)
|
|
29
|
+
- [Requirements](#-requirements)
|
|
30
|
+
- [FAQ](#-faq)
|
|
31
|
+
- [Related Links](#-related-links)
|
|
32
|
+
- [License](#-license)
|
|
33
|
+
|
|
34
|
+
---
|
|
35
|
+
|
|
36
|
+
## What is AgentAudit?
|
|
37
|
+
|
|
38
|
+
AgentAudit is a security scanner purpose-built for the AI package ecosystem. It works in two modes:
|
|
39
|
+
|
|
40
|
+
1. **CLI tool** โ Run `agentaudit` in your terminal to discover and scan MCP servers installed in your AI editors
|
|
41
|
+
2. **MCP server** โ Add to Claude Desktop, Cursor, or Windsurf so your AI agent can audit packages on your behalf
|
|
42
|
+
|
|
43
|
+
It checks packages against the [AgentAudit Trust Registry](https://agentaudit.dev) โ a shared, community-driven database of security findings โ and can perform local scans ranging from fast regex analysis to deep LLM-powered 3-pass audits.
|
|
44
|
+
|
|
45
|
+
---
|
|
46
|
+
|
|
47
|
+
## ๐ Quick Start
|
|
48
|
+
|
|
49
|
+
<p align="center">
|
|
50
|
+
<img src="docs/cli-screenshot.png" alt="AgentAudit CLI โ discover and scan" width="700">
|
|
51
|
+
</p>
|
|
52
|
+
|
|
53
|
+
### Option A: CLI (recommended)
|
|
54
|
+
|
|
55
|
+
```bash
|
|
56
|
+
# Install globally (or use npx agentaudit)
|
|
57
|
+
npm install -g agentaudit
|
|
58
|
+
|
|
59
|
+
# Discover MCP servers configured in your AI editors
|
|
60
|
+
agentaudit
|
|
61
|
+
|
|
62
|
+
# Quick scan โ clones repo, checks code with regex patterns (~2s)
|
|
63
|
+
agentaudit scan https://github.com/owner/repo
|
|
64
|
+
|
|
65
|
+
# Deep audit โ clones repo, sends code to LLM for 3-pass analysis (~30s)
|
|
66
|
+
agentaudit audit https://github.com/owner/repo
|
|
67
|
+
|
|
68
|
+
# Registry lookup โ check if a package has been audited before (no cloning)
|
|
69
|
+
agentaudit lookup fastmcp
|
|
70
|
+
```
|
|
71
|
+
|
|
72
|
+
**Example output:**
|
|
73
|
+
```
|
|
74
|
+
AgentAudit v3.9.8
|
|
75
|
+
Security scanner for AI packages
|
|
76
|
+
|
|
77
|
+
Discovering MCP servers in your AI editors...
|
|
78
|
+
|
|
79
|
+
โข Scanning Cursor ~/.cursor/mcp.json found 3 servers
|
|
80
|
+
|
|
81
|
+
โโโ tool supabase-mcp โ ok
|
|
82
|
+
โ SAFE Risk 0 https://agentaudit.dev/skills/supabase-mcp
|
|
83
|
+
โโโ tool browser-tools-mcp โ ok
|
|
84
|
+
โ โ not audited Run: agentaudit audit https://github.com/nichochar/browser-tools-mcp
|
|
85
|
+
โโโ tool filesystem โ ok
|
|
86
|
+
โ SAFE Risk 0 https://agentaudit.dev/skills/filesystem
|
|
87
|
+
|
|
88
|
+
Looking for general package scanning? Try `pip audit` or `npm audit`.
|
|
89
|
+
```
|
|
90
|
+
|
|
91
|
+
### Option B: MCP Server in your AI editor
|
|
92
|
+
|
|
93
|
+
Add AgentAudit as an MCP server โ your AI agent can then discover, scan, and audit packages using its own LLM. **No extra API key needed.**
|
|
94
|
+
|
|
95
|
+
<details>
|
|
96
|
+
<summary><strong>Claude Desktop</strong> โ <code>~/.claude/mcp.json</code></summary>
|
|
97
|
+
|
|
98
|
+
```json
|
|
99
|
+
{
|
|
100
|
+
"mcpServers": {
|
|
101
|
+
"agentaudit": {
|
|
102
|
+
"command": "npx",
|
|
103
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
104
|
+
}
|
|
105
|
+
}
|
|
106
|
+
}
|
|
107
|
+
```
|
|
108
|
+
</details>
|
|
109
|
+
|
|
110
|
+
<details>
|
|
111
|
+
<summary><strong>Cursor</strong> โ <code>.cursor/mcp.json</code> (project) or <code>~/.cursor/mcp.json</code> (global)</summary>
|
|
112
|
+
|
|
113
|
+
```json
|
|
114
|
+
{
|
|
115
|
+
"mcpServers": {
|
|
116
|
+
"agentaudit": {
|
|
117
|
+
"command": "npx",
|
|
118
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
119
|
+
}
|
|
120
|
+
}
|
|
121
|
+
}
|
|
122
|
+
```
|
|
123
|
+
</details>
|
|
124
|
+
|
|
125
|
+
<details>
|
|
126
|
+
<summary><strong>Windsurf</strong> โ <code>~/.codeium/windsurf/mcp_config.json</code></summary>
|
|
127
|
+
|
|
128
|
+
```json
|
|
129
|
+
{
|
|
130
|
+
"mcpServers": {
|
|
131
|
+
"agentaudit": {
|
|
132
|
+
"command": "npx",
|
|
133
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
134
|
+
}
|
|
135
|
+
}
|
|
136
|
+
}
|
|
137
|
+
```
|
|
138
|
+
</details>
|
|
139
|
+
|
|
140
|
+
<details>
|
|
141
|
+
<summary><strong>VS Code</strong> โ <code>.vscode/mcp.json</code></summary>
|
|
142
|
+
|
|
143
|
+
```json
|
|
144
|
+
{
|
|
145
|
+
"servers": {
|
|
146
|
+
"agentaudit": {
|
|
147
|
+
"command": "npx",
|
|
148
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
149
|
+
}
|
|
150
|
+
}
|
|
151
|
+
}
|
|
152
|
+
```
|
|
153
|
+
</details>
|
|
154
|
+
|
|
155
|
+
<details>
|
|
156
|
+
<summary><strong>Continue.dev</strong> โ <code>~/.continue/config.json</code></summary>
|
|
157
|
+
|
|
158
|
+
Add to the `mcpServers` section of your existing config:
|
|
159
|
+
```json
|
|
160
|
+
{
|
|
161
|
+
"mcpServers": [
|
|
162
|
+
{
|
|
163
|
+
"name": "agentaudit",
|
|
164
|
+
"command": "npx",
|
|
165
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
166
|
+
}
|
|
167
|
+
]
|
|
168
|
+
}
|
|
169
|
+
```
|
|
170
|
+
</details>
|
|
171
|
+
|
|
172
|
+
<details>
|
|
173
|
+
<summary><strong>Zed</strong> โ <code>~/.config/zed/settings.json</code></summary>
|
|
174
|
+
|
|
175
|
+
```json
|
|
176
|
+
{
|
|
177
|
+
"context_servers": {
|
|
178
|
+
"agentaudit": {
|
|
179
|
+
"command": {
|
|
180
|
+
"path": "npx",
|
|
181
|
+
"args": ["-y", "agentaudit", "--stdio"]
|
|
182
|
+
}
|
|
183
|
+
}
|
|
184
|
+
}
|
|
185
|
+
}
|
|
186
|
+
```
|
|
187
|
+
</details>
|
|
188
|
+
|
|
189
|
+
Then ask your agent: *"Check which MCP servers I have installed and audit any unaudited ones."*
|
|
190
|
+
|
|
191
|
+
---
|
|
192
|
+
|
|
193
|
+
## ๐ Commands Reference
|
|
194
|
+
|
|
195
|
+
| Command | Description | Example |
|
|
196
|
+
|---------|-------------|---------|
|
|
197
|
+
| `agentaudit` | Discover MCP servers (default, same as `discover`) | `agentaudit` |
|
|
198
|
+
| `agentaudit discover` | Find MCP servers in Cursor, Claude, VS Code, Windsurf | `agentaudit discover` |
|
|
199
|
+
| `agentaudit discover --quick` | Discover + auto-scan all servers | `agentaudit discover --quick` |
|
|
200
|
+
| `agentaudit discover --deep` | Discover + interactively select servers to deep-audit | `agentaudit discover --deep` |
|
|
201
|
+
| `agentaudit scan <url>` | Quick regex-based static scan (~2s) | `agentaudit scan https://github.com/owner/repo` |
|
|
202
|
+
| `agentaudit scan <url> --deep` | Deep audit (same as `audit`) | `agentaudit scan https://github.com/owner/repo --deep` |
|
|
203
|
+
| `agentaudit audit <url>` | Deep LLM-powered 3-pass audit (~30s) | `agentaudit audit https://github.com/owner/repo` |
|
|
204
|
+
| `agentaudit lookup <name>` | Look up package in trust registry | `agentaudit lookup fastmcp` |
|
|
205
|
+
| `agentaudit setup` | Register agent + configure API key | `agentaudit setup` |
|
|
206
|
+
|
|
207
|
+
### Global Flags
|
|
208
|
+
|
|
209
|
+
| Flag | Description |
|
|
210
|
+
|------|-------------|
|
|
211
|
+
| `--json` | Output machine-readable JSON to stdout |
|
|
212
|
+
| `--quiet` / `-q` | Suppress banner and decorative output (show findings only) |
|
|
213
|
+
| `--no-color` | Disable ANSI colors (also respects `NO_COLOR` env var) |
|
|
214
|
+
| `--help` / `-h` | Show help text |
|
|
215
|
+
| `-v` / `--version` | Show version |
|
|
216
|
+
|
|
217
|
+
### Exit Codes
|
|
218
|
+
|
|
219
|
+
| Code | Meaning |
|
|
220
|
+
|------|---------|
|
|
221
|
+
| `0` | Clean โ no findings detected, or successful lookup |
|
|
222
|
+
| `1` | Findings detected |
|
|
223
|
+
| `2` | Error (clone failed, network error, invalid args) |
|
|
224
|
+
|
|
225
|
+
---
|
|
226
|
+
|
|
227
|
+
## โ๏ธ Quick Scan vs Deep Audit
|
|
228
|
+
|
|
229
|
+
| | Quick Scan (`scan`) | Deep Audit (`audit`) |
|
|
230
|
+
|---|---------------------|---------------------|
|
|
231
|
+
| **Speed** | ~2 seconds | ~30 seconds |
|
|
232
|
+
| **Method** | Regex pattern matching | LLM-powered 3-pass analysis |
|
|
233
|
+
| **API key needed** | No | Yes (`ANTHROPIC_API_KEY` or `OPENAI_API_KEY`) |
|
|
234
|
+
| **False positives** | Higher (regex limitations) | Very low (context-aware) |
|
|
235
|
+
| **Detects** | Common patterns (injection, secrets, eval) | Complex attack chains, AI-specific threats, obfuscation |
|
|
236
|
+
| **Best for** | Quick triage, CI pipelines | Critical packages, pre-production review |
|
|
237
|
+
|
|
238
|
+
**Tip:** Use `agentaudit scan <url> --deep` to run a deep audit via the scan command.
|
|
239
|
+
|
|
240
|
+
---
|
|
241
|
+
|
|
242
|
+
## ๐ MCP Server
|
|
243
|
+
|
|
244
|
+
When running as an MCP server, AgentAudit exposes the following tools to your AI agent:
|
|
245
|
+
|
|
246
|
+
| Tool | Description |
|
|
247
|
+
|------|-------------|
|
|
248
|
+
| `audit_package` | Deep LLM-powered audit of a repository |
|
|
249
|
+
| `check_registry` | Look up a package in the trust registry |
|
|
250
|
+
| `submit_report` | Upload audit findings to the registry |
|
|
251
|
+
| `discover_servers` | Find MCP servers in local editor configs |
|
|
252
|
+
|
|
253
|
+
### Workflow
|
|
254
|
+
|
|
255
|
+
```
|
|
256
|
+
User asks agent to install a package
|
|
257
|
+
โ
|
|
258
|
+
โผ
|
|
259
|
+
Agent calls check_registry(package_name)
|
|
260
|
+
โ
|
|
261
|
+
โโโโโโดโโโโโ
|
|
262
|
+
โ โ
|
|
263
|
+
Found Not Found
|
|
264
|
+
โ โ
|
|
265
|
+
โผ โผ
|
|
266
|
+
Return Agent calls audit_package(repo_url)
|
|
267
|
+
score โ
|
|
268
|
+
โผ
|
|
269
|
+
LLM analyzes code (3-pass)
|
|
270
|
+
โ
|
|
271
|
+
โผ
|
|
272
|
+
Agent calls submit_report(findings)
|
|
273
|
+
โ
|
|
274
|
+
โผ
|
|
275
|
+
Return findings + risk score
|
|
276
|
+
```
|
|
277
|
+
|
|
278
|
+
---
|
|
279
|
+
|
|
280
|
+
## ๐ฏ What It Detects
|
|
281
|
+
|
|
282
|
+
<table>
|
|
283
|
+
<tr>
|
|
284
|
+
<td>
|
|
285
|
+
|
|
286
|
+
**Core Security**
|
|
287
|
+
|
|
288
|
+
|
|
289
|
+
|
|
290
|
+
|
|
291
|
+
|
|
292
|
+
|
|
293
|
+
|
|
294
|
+
|
|
295
|
+
</td>
|
|
296
|
+
<td>
|
|
297
|
+
|
|
298
|
+
**AI-Specific**
|
|
299
|
+
|
|
300
|
+
|
|
301
|
+
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
|
|
305
|
+
|
|
306
|
+
|
|
307
|
+
</td>
|
|
308
|
+
</tr>
|
|
309
|
+
<tr>
|
|
310
|
+
<td>
|
|
311
|
+
|
|
312
|
+
**MCP-Specific**
|
|
313
|
+
|
|
314
|
+
|
|
315
|
+
|
|
316
|
+
|
|
317
|
+
|
|
318
|
+
|
|
319
|
+
|
|
320
|
+
</td>
|
|
321
|
+
<td>
|
|
322
|
+
|
|
323
|
+
**Persistence & Obfuscation**
|
|
324
|
+
|
|
325
|
+
|
|
326
|
+
|
|
327
|
+
|
|
328
|
+
|
|
329
|
+
|
|
330
|
+
|
|
331
|
+
|
|
332
|
+
</td>
|
|
333
|
+
</tr>
|
|
334
|
+
</table>
|
|
335
|
+
|
|
336
|
+
---
|
|
337
|
+
|
|
338
|
+
## ๐ง How the 3-Pass Audit Works
|
|
339
|
+
|
|
340
|
+
The deep audit (`agentaudit audit`) uses a structured 3-phase LLM analysis โ not a single-shot prompt, but a rigorous multi-pass process:
|
|
341
|
+
|
|
342
|
+
| Phase | Name | What Happens |
|
|
343
|
+
|-------|------|-------------|
|
|
344
|
+
| **1** | ๐ **UNDERSTAND** | Read all files and build a **Package Profile**: purpose, category, expected behaviors, trust boundaries. No scanning yet โ the goal is to understand what the package *should* do before looking for what it *shouldn't*. |
|
|
345
|
+
| **2** | ๐ฏ **DETECT** | Evidence collection against **50+ detection patterns** across 8 categories (AI-specific, MCP, persistence, obfuscation, cross-file correlation). Only facts are recorded โ no severity judgments yet. |
|
|
346
|
+
| **3** | โ๏ธ **CLASSIFY** | Every finding goes through a **Mandatory Self-Check** (5 questions), **Exploitability Assessment**, and **Confidence Gating**. HIGH/CRITICAL findings must survive a **Devil's Advocate** challenge and include a full **Reasoning Chain**. |
|
|
347
|
+
|
|
348
|
+
**Why 3 passes?** Single-pass analysis is the #1 cause of false positives. By separating understanding โ detection โ classification:
|
|
349
|
+
|
|
350
|
+
- Phase 1 prevents flagging core functionality as suspicious (e.g., SQL execution in a database tool)
|
|
351
|
+
- Phase 2 ensures evidence is collected without severity bias
|
|
352
|
+
- Phase 3 catches false positives before they reach the report
|
|
353
|
+
|
|
354
|
+
This architecture achieved **0% false positives** on our 11-package test set, down from 42% in v2.
|
|
355
|
+
|
|
356
|
+
---
|
|
357
|
+
|
|
358
|
+
## ๐ CI/CD Integration
|
|
359
|
+
|
|
360
|
+
AgentAudit is designed for CI pipelines with proper exit codes and JSON output:
|
|
361
|
+
|
|
362
|
+
```yaml
|
|
363
|
+
# GitHub Actions example
|
|
364
|
+
- name: Scan MCP servers
|
|
365
|
+
run: |
|
|
366
|
+
npx agentaudit scan https://github.com/org/mcp-server --json --quiet > results.json
|
|
367
|
+
# Exit code 1 = findings detected โ fail the build
|
|
368
|
+
```
|
|
369
|
+
|
|
370
|
+
```bash
|
|
371
|
+
# Shell scripting
|
|
372
|
+
agentaudit scan https://github.com/owner/repo --json --quiet 2>/dev/null
|
|
373
|
+
if [ $? -eq 1 ]; then
|
|
374
|
+
echo "Security findings detected!"
|
|
375
|
+
exit 1
|
|
376
|
+
fi
|
|
377
|
+
```
|
|
378
|
+
|
|
379
|
+
### JSON Output Examples
|
|
380
|
+
|
|
381
|
+
```bash
|
|
382
|
+
# Scan with JSON output
|
|
383
|
+
agentaudit scan https://github.com/owner/repo --json
|
|
384
|
+
```
|
|
385
|
+
|
|
386
|
+
```json
|
|
387
|
+
{
|
|
388
|
+
"slug": "repo",
|
|
389
|
+
"url": "https://github.com/owner/repo",
|
|
390
|
+
"findings": [
|
|
391
|
+
{
|
|
392
|
+
"severity": "high",
|
|
393
|
+
"title": "Command injection risk",
|
|
394
|
+
"file": "src/handler.js",
|
|
395
|
+
"line": 42,
|
|
396
|
+
"snippet": "exec(`git ${userInput}`)"
|
|
397
|
+
}
|
|
398
|
+
],
|
|
399
|
+
"fileCount": 15,
|
|
400
|
+
"duration": "1.8s"
|
|
401
|
+
}
|
|
402
|
+
```
|
|
403
|
+
|
|
404
|
+
```bash
|
|
405
|
+
# Registry lookup with JSON
|
|
406
|
+
agentaudit lookup fastmcp --json
|
|
407
|
+
```
|
|
408
|
+
|
|
409
|
+
> **Coming soon:** `--fail-on <severity>` flag to set minimum severity threshold for non-zero exit (e.g., `--fail-on high` ignores low/medium findings).
|
|
410
|
+
|
|
411
|
+
---
|
|
412
|
+
|
|
413
|
+
## โ๏ธ Configuration
|
|
414
|
+
|
|
415
|
+
### Credentials
|
|
416
|
+
|
|
417
|
+
AgentAudit stores credentials in `~/.config/agentaudit/credentials.json` (or `$XDG_CONFIG_HOME/agentaudit/credentials.json`).
|
|
418
|
+
|
|
419
|
+
Run `agentaudit setup` to configure interactively, or set via environment:
|
|
420
|
+
|
|
421
|
+
```bash
|
|
422
|
+
export AGENTAUDIT_API_KEY=asf_your_key_here
|
|
423
|
+
```
|
|
424
|
+
|
|
425
|
+
### Environment Variables
|
|
426
|
+
|
|
427
|
+
| Variable | Description |
|
|
428
|
+
|----------|-------------|
|
|
429
|
+
| `AGENTAUDIT_API_KEY` | API key for registry access |
|
|
430
|
+
| `ANTHROPIC_API_KEY` | Anthropic API key for deep audits (Claude) |
|
|
431
|
+
| `OPENAI_API_KEY` | OpenAI API key for deep audits (GPT-4o) |
|
|
432
|
+
| `NO_COLOR` | Disable ANSI colors ([no-color.org](https://no-color.org)) |
|
|
433
|
+
|
|
434
|
+
---
|
|
435
|
+
|
|
436
|
+
## ๐ฆ Requirements
|
|
437
|
+
|
|
438
|
+
- **Node.js** โฅ 18.0.0
|
|
439
|
+
- **Git** (for cloning repositories during scan/audit)
|
|
440
|
+
|
|
441
|
+
---
|
|
442
|
+
|
|
443
|
+
## โ FAQ
|
|
444
|
+
|
|
445
|
+
### How do I set up AgentAudit?
|
|
446
|
+
|
|
447
|
+
```bash
|
|
448
|
+
npm install -g agentaudit
|
|
449
|
+
agentaudit setup
|
|
450
|
+
```
|
|
451
|
+
|
|
452
|
+
Or use without installing: `npx agentaudit`
|
|
453
|
+
|
|
454
|
+
### Do I need an API key?
|
|
455
|
+
|
|
456
|
+
- **Quick scan** (`scan`): No API key needed โ runs locally with regex
|
|
457
|
+
- **Deep audit** (`audit`): Needs `ANTHROPIC_API_KEY` or `OPENAI_API_KEY`
|
|
458
|
+
- **Registry lookup** (`lookup`): No key needed for reading; key needed for uploading reports
|
|
459
|
+
- **MCP server**: No extra key needed โ uses the host editor's LLM
|
|
460
|
+
|
|
461
|
+
### What data is sent externally?
|
|
462
|
+
|
|
463
|
+
- **Registry lookups**: Package name/slug is sent to `agentaudit.dev` to check for existing audits
|
|
464
|
+
- **Report uploads**: Audit findings are uploaded to the public registry (requires API key)
|
|
465
|
+
- **Deep audits**: Source code is sent to Anthropic or OpenAI for LLM analysis
|
|
466
|
+
- **Quick scans**: Everything stays local โ no data leaves your machine
|
|
467
|
+
|
|
468
|
+
### Can I use it offline?
|
|
469
|
+
|
|
470
|
+
Quick scans (`agentaudit scan`) work fully offline after cloning. Registry lookups and deep audits require network access.
|
|
471
|
+
|
|
472
|
+
### Can I use it as an MCP server without the CLI?
|
|
473
|
+
|
|
474
|
+
Yes! `npx agentaudit` starts the MCP server when invoked by an editor. The CLI and MCP server are the same package โ behavior is determined by how it's called.
|
|
475
|
+
|
|
476
|
+
### How does `discover` know which editors I use?
|
|
477
|
+
|
|
478
|
+
It checks standard config file locations for Claude Desktop, Cursor, VS Code, and Windsurf. It also checks the current working directory for project-level `.cursor/mcp.json` and `.vscode/mcp.json`.
|
|
479
|
+
|
|
480
|
+
---
|
|
481
|
+
|
|
482
|
+
## ๐ Related Links
|
|
483
|
+
|
|
484
|
+
- **Trust Registry**: [agentaudit.dev](https://agentaudit.dev)
|
|
485
|
+
- **Leaderboard**: [agentaudit.dev/leaderboard](https://agentaudit.dev/leaderboard)
|
|
486
|
+
- **Agent Skill**: [github.com/starbuck100/agentaudit-skill](https://github.com/starbuck100/agentaudit-skill) โ Full agent skill with pre-install security gate, detection patterns & peer review system
|
|
487
|
+
- **MCP Server Repository**: [github.com/starbuck100/agentaudit-mcp](https://github.com/starbuck100/agentaudit-mcp)
|
|
488
|
+
- **Report Issues**: [GitHub Issues](https://github.com/starbuck100/agentaudit-mcp/issues)
|
|
489
|
+
|
|
490
|
+
---
|
|
491
|
+
|
|
492
|
+
## ๐ License
|
|
493
|
+
|
|
494
|
+
[AGPL-3.0](LICENSE) โ Free for open source use. Commercial license available for proprietary integrations.
|
|
495
|
+
|
|
496
|
+
---
|
|
497
|
+
|
|
498
|
+
<div align="center">
|
|
499
|
+
|
|
500
|
+
**Protect your AI stack. Scan before you trust.**
|
|
501
|
+
|
|
502
|
+
[Trust Registry](https://agentaudit.dev) ยท [Leaderboard](https://agentaudit.dev/leaderboard) ยท [Report Issues](https://github.com/starbuck100/agentaudit-mcp/issues)
|
|
503
|
+
|
|
504
|
+
</div>
|