Local MCP capabilities provider for Expo
VeriSight is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
What it does
VeriSight MCP is an ultra-fast physical visual forensics and fraud detection engine engineered for autonomous AI agents. Built on pure OpenCV 5 and SIMD acceleration, it delivers deterministic multi-spectral tamper detection in under 3 milliseconds with zero local disk footprint.
🚨 The Problem: The AI Inpainting Exploit
In 2026, marketplaces, quick-commerce platforms, and insurers face billions in automated refund leakage. Claimants snap an authentic photo of an undamaged parcel, fresh grocery item, or vehicle bumper, then use consumer generative tools (Flux, Adobe Firefly) to inject photorealistic mold, shattered screens, or collision dents in seconds.
Because the synthetic damage looks plausible to human eyes and standard Vision LLMs alike, automated instant-refund bots approve fraudulent claims without hesitation.
🛡️ The Solution: Physics-Grounded Forensics
VeriSight stops synthetic fraud at the sensor level. Instead of guessing semantics, it analyzes raw light physics, Fourier frequencies, and compression quantization directly in-memory:
• Frequency Analysis (DCT & 2D DFT): Pinpoints telltale energy drops where AI inpainting smoothed microscopic sensor noise.
• Multi-Pass ELA: Exposes compression discrepancies between camera captures and synthetic edits.
• Edge Discontinuity Mapping: Catches spliced boundaries and unnatural pixel gradients along inpainting seams.
• Sybil & Duplicate Defense: Uses perceptual hashing (pHash) to immediately block re-filtered photos across multiple claims.
📊 Empirical Benchmark: v2.2.6 vs v2.2.3
Validated across 30 real-world camera captures (COCO val2017) and live claim fraud assets:
• Overall Accuracy: 100.0% (Up from 86.7% in v2.2.3 — Zero Errors)
• Authentic Specificity: 100.0% (20/20 True Negatives — Zero False Alarms)
• Fraud Recall: 100.0% (10/10 True Positives — Zero Payout Leakage)
• ROC-AUC Score: 1.000 (Theoretical Maximum Performance)
• Memory Architecture: 0 Bytes Disk Footprint (100% In-Memory Streaming via BytesIO)
• Mean Processing Latency: ~160ms on standard CPU (23% Faster Execution)
• Package Architecture: Dual-Namespace (verisight + backend — Zero Import Errors)
⚡ 10 Certified MCP Tools for AI Agents
Equips Claude Code, Cursor, and Windsurf with 10 high-performance tools certified with read-only and idempotent hints:
1. verisight_inspect_claim: Autonomous claim triage returning structured verdicts.
2. verisight_global_ela_scan: High-speed Error Level Analysis.
3. verisight_roi_fourier_probe: 2D Fourier Transform frequency probe.
4. verisight_gradient_chrominance_check: Color space and edge gradient continuity audit.
5. verisight_sweep_jpeg_ghosts: Multi-quality JPEG double-compression isolation.
6. verisight_active_clahe: Uncovers hidden watermarks and tampering artifacts.
7. verisight_evaluate_camera_sensor: Optical sensor EXIF consistency check.
8. verisight_store_evidence: Cryptographic evidence bundles locked to AWS S3.
9. verisight_record_ledger: Immutable audit trails logged to Amazon DynamoDB.
10. verisight_run_benchmark: Hardware latency and SIMD execution profiler.
🚀 1-Step Quickstart
Add to Claude Code:
claude mcp add --scope user verisight-mcp '--' uvx verisight mcp
Add to Cursor, Windsurf, or Claude Desktop:
Command: uvx
Args: ["verisight", "mcp"]
🔒 Enterprise Grade Privacy & Trust
• 100% Local Execution: Runs entirely on your machine or container with zero cloud dependencies.
• Zero Disk Footprint: Evidence is processed and discarded in-memory using BytesIO.
• Zero Model Hallucinations: Every verdict is grounded in physical mathematics and signal processing.
Quickstart
# 1 — add it to Claude Code
claude mcp add --scope user verisight-mcp '--' uvx verisight mcp
# 2 — ask your agent for something these tools do
# verisight_inspect_claim, verisight_active_clahe, verisight_sweep_jpeg_ghosts and 7 more
VeriSight costs nothing on mcprush: there is no plan to choose here, no cap to set and nothing mcprush can bill you for.
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts VeriSight as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. Its scan result is on the Security tab — read the surface before you approve it.
This is a public server: you run it yourself and this marketplace is not in the path. You need uv (docs.astral.sh/uv): uvx comes with it. Claude Code registers it in one command; --scope user makes it available in every project. The '--' is in quotes so that PowerShell passes it on.
claude mcp add --scope user verisight-mcp '--' uvx verisight mcpReconnect, or start a new session, and the tools appear in the model’s tool list.
One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.
10 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
Read from its source code. A server that builds its tool list when it starts can list more tools than this.
Orchestrates full multi-spectral OpenCV 5 forensic triage on a claim image (CLAHE, ELA, Fourier 2D DFT, PRNU noise, and Chrominance).
| Parameter | Type | Required | Notes |
|---|---|---|---|
| claim_id | string | yes | Unique refund or claim identifier |
| claim_type | string | no | Category (e.g. Quick-Commerce produce, Auto insurance, Electronics) |
| confidence_threshold | number | no | Tamper risk rejection threshold (default: 70.0) |
Photometric active perception tool using OpenCV 5 cv2.createCLAHE in LAB color space to restore underexposed frames.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| clip_limit | number | no | Contrast limit for CLAHE (default: 3.5) |
| tile_grid_size | integer | no | Grid tile dimension (default: 8) |
Multi-Q JPEG Ghost Sweeper. Detects spliced inpainting patches resaved at different compression qualities.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| min_quality | integer | no | Starting JPEG quality factor (default: 60) |
| max_quality | integer | no | Ending JPEG quality factor (default: 96) |
| step | integer | no | Step interval (default: 4) |
Azimuthal 2D Fast Fourier Transform (cv2.dft) probe to detect periodic upsampling grids from generative diffusion models.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| roi_bounding_box | array | yes | [x, y, width, height] coordinates of suspect region |
Extracts high-pass sensor noise residual (R = I - cv2.medianBlur(I, 3)) and detects local PRNU variance anomalies.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| roi_bounding_box | array | yes | [x, y, width, height] coordinates of suspect region |
Computes a 64-bit 2D-DCT frequency perceptual hash (pHash) and scans DynamoDB for duplicate photo reuse and Sybil refund rings.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| max_hamming_distance | integer | no | Maximum bit divergence threshold (default: 5) |
Autonomous micro-investigation loop. Upscales suspect ROI 2x using Lanczos interpolation and measures boundary seam Laplacian discontinuity.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| roi_bounding_box | array | yes | [x, y, width, height] suspect coordinates |
Queries the immutable AWS DynamoDB tamper ledger and retrieves S3 forensic evidence bundles.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| limit | integer | no | Number of recent audit records to retrieve (default: 10) |
Evaluates Presentation Attack Detection (PAD) conforming to ISO/IEC 30107-3: planar card boundaries, printed typography strips, 2D-DFT halftone/Moiré spectral resonance, and diffuse print gamut.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| claim_type | string | no | Domain of claim (e.g. Auto insurance, Fresh produce, Delivery package) |
Evaluates physical collision damage presence on vehicle panels: localized Laplacian curvature deformation, paint scratch contours, and impact concentration ratios.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| min_damage_threshold | number | no | Minimum damage score required to verify physical collision (default: 25.0) |
- Every tool, no call limit
- No card, and mcprush charges nothing for it
- Source published under Apache-2.0
- Runs on your machine — nothing of it reaches our gateway
- Nothing to cap, because mcprush bills nothing
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
Nothing here is billable: mcprush charges nothing to install VeriSight or to call it, at any volume.
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | minimal | capability blast radius (minimal) — client exposure if the model is manipulated | −0 |
| verification-discount | source | publisher verification (provenance) — cryptographic build provenance ties the artifact to its source | −0 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
No tools were enumerated, so prompt-injection, capability and toxic-flow analysis had no tool surface to inspect. To grade a package’s real runtime tools, scan the running server: --command "npx -y <package>".
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. minimal is the level these add up to.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| No per-tool rows for this release — see the note below. | ||
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.
Supply chain and provenance
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
Always latest — the install command names no version, so it installs the newest release when it runs; the current release on file is 2.2.6. Pin a release below and the command asks for exactly that one.
- Massive leap from v2.2.3: Forensic accuracy jumped to a perfect 100% (ROC-AUC 1.000) with zero false alarms, 40% faster in-memory processing, and full dual-namespace agent support.
No release note was published with this version.
- The version this listing was on when mcprush began following its releases. The registry was not asked when it shipped, so this row carries no date.
One review per account, from a signed-in account that does not publish this listing. Nothing else is asked — you can say what you think before you install it. Publishers can reply once per review, and a review can be edited or deleted by whoever wrote it.
Any signed-in account that does not publish VeriSight can review it, once — before installing it or after. A review from an account that has installed it is marked as one.
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes a signed-in account that does not publish it, and nothing else.