verisight 2.2.1__tar.gz → 2.2.3__tar.gz
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.
- {verisight-2.2.1/verisight.egg-info → verisight-2.2.3}/PKG-INFO +30 -2
- {verisight-2.2.1 → verisight-2.2.3}/README.md +29 -1
- {verisight-2.2.1 → verisight-2.2.3}/backend/__init__.py +1 -1
- {verisight-2.2.1 → verisight-2.2.3}/backend/agent.py +49 -27
- {verisight-2.2.1 → verisight-2.2.3}/backend/main.py +4 -4
- {verisight-2.2.1 → verisight-2.2.3}/backend/mcp_server.py +2 -2
- {verisight-2.2.1 → verisight-2.2.3}/pyproject.toml +1 -1
- {verisight-2.2.1 → verisight-2.2.3/verisight.egg-info}/PKG-INFO +30 -2
- {verisight-2.2.1 → verisight-2.2.3}/LICENSE +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/aws_adapter.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/benchmark_runner.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/cli.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/forensics.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/generate_demo_samples.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/generate_real_life_samples.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/test_mcp_tools.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/backend/test_sanity.py +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/setup.cfg +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/verisight.egg-info/SOURCES.txt +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/verisight.egg-info/dependency_links.txt +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/verisight.egg-info/entry_points.txt +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/verisight.egg-info/requires.txt +0 -0
- {verisight-2.2.1 → verisight-2.2.3}/verisight.egg-info/top_level.txt +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: verisight
|
|
3
|
-
Version: 2.2.
|
|
3
|
+
Version: 2.2.3
|
|
4
4
|
Summary: Sub-3ms Physical Multi-Spectral Image Forensics & Presentation Attack Detection engine powered by OpenCV 5 and Model Context Protocol (MCP).
|
|
5
5
|
Author-email: Aditya Raj Tiwari <adityasantosh2026@gmail.com>
|
|
6
6
|
License-Expression: Apache-2.0
|
|
@@ -34,6 +34,8 @@ Dynamic: license-file
|
|
|
34
34
|
# 🔬 VeriSight MCP — Autonomous Physical AI Visual Forensics Engine
|
|
35
35
|
|
|
36
36
|
[](https://github.com/Adityarajtiwari880/verisight/actions/workflows/ci.yml)
|
|
37
|
+
[](https://pypi.org/project/verisight/)
|
|
38
|
+
[](https://mcprush.com/to/verisight-mcp?ref=badge)
|
|
37
39
|
[](https://m8ven.ai/mcp/adityarajtiwari880-verisight-11ohzr?s=readme)
|
|
38
40
|
[](LICENSE)
|
|
39
41
|
[](pyproject.toml)
|
|
@@ -43,6 +45,10 @@ Dynamic: license-file
|
|
|
43
45
|
[](https://nextjs.org/)
|
|
44
46
|
[](https://fastapi.tiangolo.com/)
|
|
45
47
|
|
|
48
|
+
<p align="center">
|
|
49
|
+
<a href="https://mcprush.com/to/verisight-mcp?ref=badge"><img width="380" height="200" src="https://mcprush.com/badge/verisight-mcp/card.svg" alt="VeriSight MCP server" /></a>
|
|
50
|
+
</p>
|
|
51
|
+
|
|
46
52
|
> **Sub-3ms Physical AI and multi-spectral visual forensics engine built on pure OpenCV 5 and SIMD acceleration.** Detects generative AI inpainting, deepfake manipulation, duplicate photo reuse, and presentation recapture attacks with zero heavyweight neural network dependencies.
|
|
47
53
|
|
|
48
54
|
---
|
|
@@ -77,6 +83,23 @@ Authentic Image ──> Generative AI Inpaint (Flux / Firefly) ──> Instant R
|
|
|
77
83
|
| **Generalization & Drift** | Vulnerable to adversarial attacks & new diffusion models | **Grounded in mathematical physical optics & sensor physics** |
|
|
78
84
|
| **Audit Explainability** | Black-box softmax probabilities | **Deterministic mathematical proof (ELA, PRNU, 2D FFT, Ghosts)** |
|
|
79
85
|
| **Recapture Attack Defense**| None (Foiled by 2D screen or paper printout) | **Built-in Presentation Attack Detection (PAD)** |
|
|
86
|
+
| **External API Cost** | \$0.01–\$0.05/image via VLM/SaaS wrappers | **\$0.00 (100% offline/local C++ OpenCV 5)** |
|
|
87
|
+
|
|
88
|
+
---
|
|
89
|
+
|
|
90
|
+
## 🎯 Verified Real-World Evaluation Suite (`10/10` Passing in CI)
|
|
91
|
+
|
|
92
|
+
Verified automatically on every commit via [`backend/test_sanity.py`](backend/test_sanity.py) and [`backend/test_mcp_tools.py`](backend/test_mcp_tools.py):
|
|
93
|
+
|
|
94
|
+
| Test Scenario | Ground Truth | VeriSight Verdict | Risk Score |
|
|
95
|
+
| :--- | :--- | :--- | :---: |
|
|
96
|
+
| Genuine Unedited Photo | Authentic | `AUTO_APPROVE_CLAIM` | `10.9%` |
|
|
97
|
+
| Generative AI Inpainted Damage | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
98
|
+
| Genuine Digital Coupon Screenshot | Authentic | `AUTO_APPROVE_CLAIM` | `30.8%` |
|
|
99
|
+
| AI Altered Coupon Code (`D` → `C`) | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
100
|
+
| Printed Card Recapture Attack | PAD Attack | `REJECT_RECAPTURED_2D_MEDIA` | `97.5%` |
|
|
101
|
+
| Duplicate Photo Reuse (Sybil Ring) | Reuse Fraud | `REJECT_DUPLICATE_CLAIM_FRAUD` | `99.4%` |
|
|
102
|
+
| Selfie / Off-Subject in Produce or Invoice | Invalid Subject | `REJECT_INVALID_CLAIM_SUBJECT` | `Blocked` |
|
|
80
103
|
|
|
81
104
|
---
|
|
82
105
|
|
|
@@ -148,12 +171,17 @@ verisight server --port 8000
|
|
|
148
171
|
|
|
149
172
|
VeriSight natively implements the **Anthropic Model Context Protocol (MCP)** specification (`protocol_version: 2024-11-05`). External LLM agents (**Claude Desktop**, **Cursor**, **Windsurf**, **LangChain**) can discover and call VeriSight tools autonomously via high-performance standard I/O (`stdio`).
|
|
150
173
|
|
|
151
|
-
### 1-Click Zero-Install (`uvx`)
|
|
174
|
+
### 1-Click Zero-Install (`uvx` & `claude mcp`)
|
|
152
175
|
Run directly via Astral `uvx` with zero manual cloning or environment setup:
|
|
153
176
|
```bash
|
|
154
177
|
uvx verisight mcp
|
|
155
178
|
```
|
|
156
179
|
|
|
180
|
+
Or add directly to **Claude Code CLI**:
|
|
181
|
+
```bash
|
|
182
|
+
claude mcp add --scope user verisight-mcp '--' uvx verisight mcp
|
|
183
|
+
```
|
|
184
|
+
|
|
157
185
|
### Claude Desktop & Cursor Configuration
|
|
158
186
|
|
|
159
187
|
Add VeriSight directly to your `claude_desktop_config.json` (or Cursor MCP settings):
|
|
@@ -1,6 +1,8 @@
|
|
|
1
1
|
# 🔬 VeriSight MCP — Autonomous Physical AI Visual Forensics Engine
|
|
2
2
|
|
|
3
3
|
[](https://github.com/Adityarajtiwari880/verisight/actions/workflows/ci.yml)
|
|
4
|
+
[](https://pypi.org/project/verisight/)
|
|
5
|
+
[](https://mcprush.com/to/verisight-mcp?ref=badge)
|
|
4
6
|
[](https://m8ven.ai/mcp/adityarajtiwari880-verisight-11ohzr?s=readme)
|
|
5
7
|
[](LICENSE)
|
|
6
8
|
[](pyproject.toml)
|
|
@@ -10,6 +12,10 @@
|
|
|
10
12
|
[](https://nextjs.org/)
|
|
11
13
|
[](https://fastapi.tiangolo.com/)
|
|
12
14
|
|
|
15
|
+
<p align="center">
|
|
16
|
+
<a href="https://mcprush.com/to/verisight-mcp?ref=badge"><img width="380" height="200" src="https://mcprush.com/badge/verisight-mcp/card.svg" alt="VeriSight MCP server" /></a>
|
|
17
|
+
</p>
|
|
18
|
+
|
|
13
19
|
> **Sub-3ms Physical AI and multi-spectral visual forensics engine built on pure OpenCV 5 and SIMD acceleration.** Detects generative AI inpainting, deepfake manipulation, duplicate photo reuse, and presentation recapture attacks with zero heavyweight neural network dependencies.
|
|
14
20
|
|
|
15
21
|
---
|
|
@@ -44,6 +50,23 @@ Authentic Image ──> Generative AI Inpaint (Flux / Firefly) ──> Instant R
|
|
|
44
50
|
| **Generalization & Drift** | Vulnerable to adversarial attacks & new diffusion models | **Grounded in mathematical physical optics & sensor physics** |
|
|
45
51
|
| **Audit Explainability** | Black-box softmax probabilities | **Deterministic mathematical proof (ELA, PRNU, 2D FFT, Ghosts)** |
|
|
46
52
|
| **Recapture Attack Defense**| None (Foiled by 2D screen or paper printout) | **Built-in Presentation Attack Detection (PAD)** |
|
|
53
|
+
| **External API Cost** | \$0.01–\$0.05/image via VLM/SaaS wrappers | **\$0.00 (100% offline/local C++ OpenCV 5)** |
|
|
54
|
+
|
|
55
|
+
---
|
|
56
|
+
|
|
57
|
+
## 🎯 Verified Real-World Evaluation Suite (`10/10` Passing in CI)
|
|
58
|
+
|
|
59
|
+
Verified automatically on every commit via [`backend/test_sanity.py`](backend/test_sanity.py) and [`backend/test_mcp_tools.py`](backend/test_mcp_tools.py):
|
|
60
|
+
|
|
61
|
+
| Test Scenario | Ground Truth | VeriSight Verdict | Risk Score |
|
|
62
|
+
| :--- | :--- | :--- | :---: |
|
|
63
|
+
| Genuine Unedited Photo | Authentic | `AUTO_APPROVE_CLAIM` | `10.9%` |
|
|
64
|
+
| Generative AI Inpainted Damage | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
65
|
+
| Genuine Digital Coupon Screenshot | Authentic | `AUTO_APPROVE_CLAIM` | `30.8%` |
|
|
66
|
+
| AI Altered Coupon Code (`D` → `C`) | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
67
|
+
| Printed Card Recapture Attack | PAD Attack | `REJECT_RECAPTURED_2D_MEDIA` | `97.5%` |
|
|
68
|
+
| Duplicate Photo Reuse (Sybil Ring) | Reuse Fraud | `REJECT_DUPLICATE_CLAIM_FRAUD` | `99.4%` |
|
|
69
|
+
| Selfie / Off-Subject in Produce or Invoice | Invalid Subject | `REJECT_INVALID_CLAIM_SUBJECT` | `Blocked` |
|
|
47
70
|
|
|
48
71
|
---
|
|
49
72
|
|
|
@@ -115,12 +138,17 @@ verisight server --port 8000
|
|
|
115
138
|
|
|
116
139
|
VeriSight natively implements the **Anthropic Model Context Protocol (MCP)** specification (`protocol_version: 2024-11-05`). External LLM agents (**Claude Desktop**, **Cursor**, **Windsurf**, **LangChain**) can discover and call VeriSight tools autonomously via high-performance standard I/O (`stdio`).
|
|
117
140
|
|
|
118
|
-
### 1-Click Zero-Install (`uvx`)
|
|
141
|
+
### 1-Click Zero-Install (`uvx` & `claude mcp`)
|
|
119
142
|
Run directly via Astral `uvx` with zero manual cloning or environment setup:
|
|
120
143
|
```bash
|
|
121
144
|
uvx verisight mcp
|
|
122
145
|
```
|
|
123
146
|
|
|
147
|
+
Or add directly to **Claude Code CLI**:
|
|
148
|
+
```bash
|
|
149
|
+
claude mcp add --scope user verisight-mcp '--' uvx verisight mcp
|
|
150
|
+
```
|
|
151
|
+
|
|
124
152
|
### Claude Desktop & Cursor Configuration
|
|
125
153
|
|
|
126
154
|
Add VeriSight directly to your `claude_desktop_config.json` (or Cursor MCP settings):
|
|
@@ -38,7 +38,7 @@ class ForensicAgent:
|
|
|
38
38
|
JPEG Ghost Sweeper, and PRNU Sensor Noise.
|
|
39
39
|
"""
|
|
40
40
|
def __init__(self):
|
|
41
|
-
self.version = "2.2.
|
|
41
|
+
self.version = "2.2.3"
|
|
42
42
|
|
|
43
43
|
def run_investigation(
|
|
44
44
|
self,
|
|
@@ -443,51 +443,73 @@ class ForensicAgent:
|
|
|
443
443
|
is_full_frame = (rw > 0.65 * working_dim[0] and rh > 0.65 * working_dim[1])
|
|
444
444
|
has_isolated_cluster = not is_full_frame and (clusters_detected > 0) and (ela_score > 25.0)
|
|
445
445
|
|
|
446
|
-
#
|
|
447
|
-
#
|
|
448
|
-
|
|
449
|
-
|
|
450
|
-
|
|
451
|
-
|
|
446
|
+
# Natural Camera vs Document / Synthetic Domain Differentiation:
|
|
447
|
+
# In natural camera photos (damaged items, vehicles, nature, people), natural optical color transitions
|
|
448
|
+
# produce high whole-frame chrominance variance that must not be confused with generative AI fringing.
|
|
449
|
+
# In documents/vouchers, paper is predominantly monochrome/duotone so character editing creates high chrominance spikes.
|
|
450
|
+
is_doc_or_voucher = scene_triage.get("has_document_structure", False) or any(
|
|
451
|
+
k in claim_type.lower() for k in ["voucher", "coupon", "bill", "invoice", "receipt", "document", "text"]
|
|
452
452
|
)
|
|
453
453
|
|
|
454
|
-
#
|
|
455
|
-
|
|
456
|
-
|
|
457
|
-
|
|
458
|
-
|
|
459
|
-
|
|
460
|
-
|
|
461
|
-
|
|
462
|
-
|
|
454
|
+
# Generative AI detection & Corroboration:
|
|
455
|
+
if is_doc_or_voucher:
|
|
456
|
+
is_generative_ai = (
|
|
457
|
+
(cr_sobel_std >= 6.0 and ela_score >= 35.0 and has_isolated_cluster) or
|
|
458
|
+
(sensor_noise_std < 1.15 and ela_score >= 30.0) or
|
|
459
|
+
(harmonic_ratio >= 2.8 and fft_energy >= 85.0 and ela_score >= 38.0) or
|
|
460
|
+
(ghost_discrepancy >= 0.82 and noise_divergence >= 35.0 and has_isolated_cluster) or
|
|
461
|
+
(is_feathered and has_isolated_cluster)
|
|
462
|
+
)
|
|
463
|
+
has_corroboration = (
|
|
464
|
+
is_feathered or
|
|
465
|
+
(ghost_discrepancy >= 0.8) or
|
|
466
|
+
(noise_divergence >= 45.0 and ela_score >= 25.0) or
|
|
467
|
+
(ela_score >= 68.0) or
|
|
468
|
+
is_generative_ai or
|
|
469
|
+
provenance.get("is_edited_software_flag", False)
|
|
470
|
+
)
|
|
471
|
+
else:
|
|
472
|
+
is_generative_ai = (
|
|
473
|
+
(has_isolated_cluster and is_feathered and ela_score >= 35.0) or
|
|
474
|
+
(has_isolated_cluster and sensor_noise_std < 1.15 and noise_divergence >= 35.0 and ela_score >= 30.0) or
|
|
475
|
+
(harmonic_ratio >= 2.8 and fft_energy >= 85.0 and is_feathered) or
|
|
476
|
+
(has_isolated_cluster and is_feathered and (ghost_discrepancy >= 0.82 or noise_divergence >= 40.0))
|
|
477
|
+
)
|
|
478
|
+
has_corroboration = (
|
|
479
|
+
is_feathered or
|
|
480
|
+
(ghost_discrepancy >= 0.82 and has_isolated_cluster and is_feathered) or
|
|
481
|
+
(noise_divergence >= 45.0 and ela_score >= 35.0 and has_isolated_cluster and is_feathered) or
|
|
482
|
+
is_generative_ai or
|
|
483
|
+
provenance.get("is_edited_software_flag", False)
|
|
484
|
+
)
|
|
463
485
|
|
|
464
486
|
# 1. Calibrated ELA contribution:
|
|
465
487
|
# Organic cameras: ELA <= 20 (contrib 0-5%)
|
|
466
|
-
# Moderate recompression: ELA 20-35 (contrib 5-
|
|
467
|
-
# Corroborated inpainting: ELA > 35 (contrib
|
|
468
|
-
# Uncorroborated
|
|
488
|
+
# Moderate recompression: ELA 20-35 (contrib 5-27%)
|
|
489
|
+
# Corroborated inpainting: ELA > 35 (contrib 28-75%)
|
|
490
|
+
# Uncorroborated natural optical edges / texture: capped safely at 16%
|
|
469
491
|
if ela_score <= 20.0:
|
|
470
492
|
ela_contrib = max(0.0, (ela_score - 14.0) * 0.8)
|
|
471
493
|
elif ela_score <= 35.0:
|
|
472
|
-
ela_contrib = 5.0 + (ela_score - 20.0) * 1.
|
|
494
|
+
ela_contrib = 5.0 + (ela_score - 20.0) * 1.5
|
|
473
495
|
else:
|
|
474
|
-
if has_corroboration:
|
|
475
|
-
ela_contrib =
|
|
496
|
+
if has_corroboration and has_isolated_cluster:
|
|
497
|
+
ela_contrib = 28.0 + (ela_score - 35.0) * 1.8
|
|
476
498
|
else:
|
|
477
|
-
ela_contrib = min(
|
|
499
|
+
ela_contrib = min(16.0, 5.0 + (ela_score - 20.0) * 0.35)
|
|
478
500
|
|
|
479
501
|
# 2. Diffusion / Generative AI Spectral penalty:
|
|
480
502
|
diffusion_contrib = 30.0 if is_generative_ai else 0.0
|
|
481
503
|
|
|
482
504
|
# 3. Chrominance & Multi-probe contributions:
|
|
483
|
-
if has_corroboration:
|
|
484
|
-
chrom_contrib = min(
|
|
505
|
+
if has_corroboration and has_isolated_cluster:
|
|
506
|
+
chrom_contrib = min(35.0, max(0.0, (chrom_diff - 25.0) * 0.45))
|
|
485
507
|
else:
|
|
486
|
-
chrom_contrib = min(
|
|
508
|
+
chrom_contrib = min(5.0, max(0.0, (chrom_diff - 25.0) * 0.08)) if has_isolated_cluster else 0.0
|
|
487
509
|
|
|
488
510
|
noise_contrib = max(0.0, (noise_divergence - 70.0) * 0.35) if (has_isolated_cluster and has_corroboration) else 0.0
|
|
489
511
|
ghost_contrib = max(0.0, (ghost_discrepancy - 0.8) * 6.0) if (has_isolated_cluster and has_corroboration) else 0.0
|
|
490
|
-
seam_contrib =
|
|
512
|
+
seam_contrib = 25.0 if (is_feathered and has_isolated_cluster) else 0.0
|
|
491
513
|
software_contrib = 35.0 if provenance.get("is_edited_software_flag") else 0.0
|
|
492
514
|
|
|
493
515
|
# Sensitivity modifier
|
|
@@ -31,7 +31,7 @@ from mcp_server import MCP_TOOLS_MANIFEST, handle_json_rpc
|
|
|
31
31
|
app = FastAPI(
|
|
32
32
|
title="VeriSight MCP - OpenCV 5 Agentic Forensics Engine",
|
|
33
33
|
description="Multi-spectral forensic vision agent powered by OpenCV 5 on AWS Graviton Arm64",
|
|
34
|
-
version="2.2.
|
|
34
|
+
version="2.2.3"
|
|
35
35
|
)
|
|
36
36
|
|
|
37
37
|
app.add_middleware(
|
|
@@ -47,7 +47,7 @@ def root_status():
|
|
|
47
47
|
return {
|
|
48
48
|
"service": "VeriSight MCP - Autonomous Physical AI Forensics Engine",
|
|
49
49
|
"status": "ONLINE",
|
|
50
|
-
"version": "2.2.
|
|
50
|
+
"version": "2.2.3",
|
|
51
51
|
"engine": "OpenCV 5.0.0 (ARM64 SIMD)",
|
|
52
52
|
"docs_url": "/docs",
|
|
53
53
|
"api_url": "/api",
|
|
@@ -56,7 +56,7 @@ def root_status():
|
|
|
56
56
|
|
|
57
57
|
@app.get("/health")
|
|
58
58
|
def health_check():
|
|
59
|
-
return {"status": "healthy", "service": "verisight-backend", "version": "2.2.
|
|
59
|
+
return {"status": "healthy", "service": "verisight-backend", "version": "2.2.3"}
|
|
60
60
|
|
|
61
61
|
@app.get("/api")
|
|
62
62
|
@app.get("/api/")
|
|
@@ -64,7 +64,7 @@ def api_index():
|
|
|
64
64
|
return {
|
|
65
65
|
"service": "VeriSight MCP REST API",
|
|
66
66
|
"status": "ONLINE",
|
|
67
|
-
"version": "2.2.
|
|
67
|
+
"version": "2.2.3",
|
|
68
68
|
"documentation": "/docs",
|
|
69
69
|
"endpoints": {
|
|
70
70
|
"health": "/health",
|
|
@@ -251,7 +251,7 @@ def handle_json_rpc(payload: Dict[str, Any], agent_pipeline_callback) -> Optiona
|
|
|
251
251
|
},
|
|
252
252
|
"serverInfo": {
|
|
253
253
|
"name": "verisight-mcp-engine",
|
|
254
|
-
"version": "2.2.
|
|
254
|
+
"version": "2.2.3"
|
|
255
255
|
}
|
|
256
256
|
}
|
|
257
257
|
}
|
|
@@ -268,7 +268,7 @@ def handle_json_rpc(payload: Dict[str, Any], agent_pipeline_callback) -> Optiona
|
|
|
268
268
|
"protocol_version": "2024-11-05",
|
|
269
269
|
"server_info": {
|
|
270
270
|
"name": "verisight-mcp-engine",
|
|
271
|
-
"version": "2.2.
|
|
271
|
+
"version": "2.2.3",
|
|
272
272
|
"capabilities": {"tools": True, "resources": False, "prompts": False}
|
|
273
273
|
}
|
|
274
274
|
}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: verisight
|
|
3
|
-
Version: 2.2.
|
|
3
|
+
Version: 2.2.3
|
|
4
4
|
Summary: Sub-3ms Physical Multi-Spectral Image Forensics & Presentation Attack Detection engine powered by OpenCV 5 and Model Context Protocol (MCP).
|
|
5
5
|
Author-email: Aditya Raj Tiwari <adityasantosh2026@gmail.com>
|
|
6
6
|
License-Expression: Apache-2.0
|
|
@@ -34,6 +34,8 @@ Dynamic: license-file
|
|
|
34
34
|
# 🔬 VeriSight MCP — Autonomous Physical AI Visual Forensics Engine
|
|
35
35
|
|
|
36
36
|
[](https://github.com/Adityarajtiwari880/verisight/actions/workflows/ci.yml)
|
|
37
|
+
[](https://pypi.org/project/verisight/)
|
|
38
|
+
[](https://mcprush.com/to/verisight-mcp?ref=badge)
|
|
37
39
|
[](https://m8ven.ai/mcp/adityarajtiwari880-verisight-11ohzr?s=readme)
|
|
38
40
|
[](LICENSE)
|
|
39
41
|
[](pyproject.toml)
|
|
@@ -43,6 +45,10 @@ Dynamic: license-file
|
|
|
43
45
|
[](https://nextjs.org/)
|
|
44
46
|
[](https://fastapi.tiangolo.com/)
|
|
45
47
|
|
|
48
|
+
<p align="center">
|
|
49
|
+
<a href="https://mcprush.com/to/verisight-mcp?ref=badge"><img width="380" height="200" src="https://mcprush.com/badge/verisight-mcp/card.svg" alt="VeriSight MCP server" /></a>
|
|
50
|
+
</p>
|
|
51
|
+
|
|
46
52
|
> **Sub-3ms Physical AI and multi-spectral visual forensics engine built on pure OpenCV 5 and SIMD acceleration.** Detects generative AI inpainting, deepfake manipulation, duplicate photo reuse, and presentation recapture attacks with zero heavyweight neural network dependencies.
|
|
47
53
|
|
|
48
54
|
---
|
|
@@ -77,6 +83,23 @@ Authentic Image ──> Generative AI Inpaint (Flux / Firefly) ──> Instant R
|
|
|
77
83
|
| **Generalization & Drift** | Vulnerable to adversarial attacks & new diffusion models | **Grounded in mathematical physical optics & sensor physics** |
|
|
78
84
|
| **Audit Explainability** | Black-box softmax probabilities | **Deterministic mathematical proof (ELA, PRNU, 2D FFT, Ghosts)** |
|
|
79
85
|
| **Recapture Attack Defense**| None (Foiled by 2D screen or paper printout) | **Built-in Presentation Attack Detection (PAD)** |
|
|
86
|
+
| **External API Cost** | \$0.01–\$0.05/image via VLM/SaaS wrappers | **\$0.00 (100% offline/local C++ OpenCV 5)** |
|
|
87
|
+
|
|
88
|
+
---
|
|
89
|
+
|
|
90
|
+
## 🎯 Verified Real-World Evaluation Suite (`10/10` Passing in CI)
|
|
91
|
+
|
|
92
|
+
Verified automatically on every commit via [`backend/test_sanity.py`](backend/test_sanity.py) and [`backend/test_mcp_tools.py`](backend/test_mcp_tools.py):
|
|
93
|
+
|
|
94
|
+
| Test Scenario | Ground Truth | VeriSight Verdict | Risk Score |
|
|
95
|
+
| :--- | :--- | :--- | :---: |
|
|
96
|
+
| Genuine Unedited Photo | Authentic | `AUTO_APPROVE_CLAIM` | `10.9%` |
|
|
97
|
+
| Generative AI Inpainted Damage | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
98
|
+
| Genuine Digital Coupon Screenshot | Authentic | `AUTO_APPROVE_CLAIM` | `30.8%` |
|
|
99
|
+
| AI Altered Coupon Code (`D` → `C`) | Fraud | `REJECT_SYNTHETIC_TAMPER` | `99.4%` |
|
|
100
|
+
| Printed Card Recapture Attack | PAD Attack | `REJECT_RECAPTURED_2D_MEDIA` | `97.5%` |
|
|
101
|
+
| Duplicate Photo Reuse (Sybil Ring) | Reuse Fraud | `REJECT_DUPLICATE_CLAIM_FRAUD` | `99.4%` |
|
|
102
|
+
| Selfie / Off-Subject in Produce or Invoice | Invalid Subject | `REJECT_INVALID_CLAIM_SUBJECT` | `Blocked` |
|
|
80
103
|
|
|
81
104
|
---
|
|
82
105
|
|
|
@@ -148,12 +171,17 @@ verisight server --port 8000
|
|
|
148
171
|
|
|
149
172
|
VeriSight natively implements the **Anthropic Model Context Protocol (MCP)** specification (`protocol_version: 2024-11-05`). External LLM agents (**Claude Desktop**, **Cursor**, **Windsurf**, **LangChain**) can discover and call VeriSight tools autonomously via high-performance standard I/O (`stdio`).
|
|
150
173
|
|
|
151
|
-
### 1-Click Zero-Install (`uvx`)
|
|
174
|
+
### 1-Click Zero-Install (`uvx` & `claude mcp`)
|
|
152
175
|
Run directly via Astral `uvx` with zero manual cloning or environment setup:
|
|
153
176
|
```bash
|
|
154
177
|
uvx verisight mcp
|
|
155
178
|
```
|
|
156
179
|
|
|
180
|
+
Or add directly to **Claude Code CLI**:
|
|
181
|
+
```bash
|
|
182
|
+
claude mcp add --scope user verisight-mcp '--' uvx verisight mcp
|
|
183
|
+
```
|
|
184
|
+
|
|
157
185
|
### Claude Desktop & Cursor Configuration
|
|
158
186
|
|
|
159
187
|
Add VeriSight directly to your `claude_desktop_config.json` (or Cursor MCP settings):
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|