# zentric-protocol-mcp [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/ZentricProtocol/zentricprotocol  
**GitHub Stars:** 0  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/zentric-protocol-mcp

## Description
Prompt injection + PII detection for LLM apps and agent pipelines. Deterministic, sub-25ms.

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

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

## Documentation

## What zentric-protocol-mcp MCP server does

The zentric-protocol-mcp MCP server represents Zentric Protocol’s documented inspection layer for LLM applications and agent pipelines. It accepts an input signal, checks it for known prompt-injection and jailbreak patterns, optionally identifies and anonymizes personally identifiable information, and returns a structured result.

The service exposes three documented outcomes: `CLEARED` when checks pass, `BLOCKED` when an injection or high-risk pattern is found, and `ANONYMIZED` when PII is detected and redacted. Each processed request also produces a report containing identifiers, a UTC timestamp, a SHA-256 value, the verdict, integrity findings, privacy findings, and latency information.

## How it works

Requests are sent to the REST endpoint at `https://api.zentricprotocol.com/v1/analyze` with a bearer token and JSON input. The request can select the `integrity` and `privacy` modules, enable anonymization, and specify a language or use automatic language detection.

IntegrityGuard uses deterministic signature matching rather than an ML model in the verdict path. The README lists 22 injection signatures and support for English, Spanish, French, German, Portuguese, Chinese, and Japanese. PrivacyGuard recognizes 12 PII categories, including email addresses, phone numbers, SSNs, passports, IBANs, SWIFT codes, and regional identifiers. Supported handling operators include redaction, masking, tokenization, and pseudonymization.

The returned report is intended to provide a per-request audit record. The documented response includes the verdict, report, any anonymized input, and measured latency. The repository describes sub-millisecond server-side processing for the inspection path.

## Setup and configuration

Access requires an active Zentric API key. The documented request uses an `Authorization: Bearer` header and sends JSON with `input`, `modules`, and optional `options` fields. A free tier provides 10,000 requests per month without a credit card; paid tiers increase the monthly request allowance.

The README documents the REST API as available now. Python and Node.js SDK installation commands are listed as coming in Q3 2026, so they should not be treated as currently available setup methods. No executable MCP server command or standard MCP client configuration is provided in the supplied material.

## Tools and capabilities

The documented API supports:

- Inspecting prompts, user inputs, responses, RAG chunks, and tool outputs.
- Detecting catalogued prompt injections, jailbreak attempts, and instruction overrides.
- Detecting PII across regional formats and languages.
- Returning anonymized input when privacy processing is enabled.
- Producing signed audit data with UUID, SHA-256, UTC timestamp, verdict, and findings.
- Benchmarking the public implementation with `npm run benchmark` against the referenced prompt-injection dataset.

These capabilities are described as HTTP API behavior; the material does not define named MCP tools or their schemas.

## Limitations and notes

The zentric-protocol-mcp MCP server should not be evaluated as a fully local, open-source detection engine based on this repository. The README explicitly states that the processing engine, injection signature database, PII engine, model data, and audit-signing infrastructure run on Zentric’s infrastructure and require an active license. The repository instead contains authentication middleware, a Stripe webhook handler, Supabase schema and migrations, interface contracts, and documentation.

Deterministic matching only covers catalogued signatures, so the stated precision and zero-false-positive claims apply to known patterns rather than every possible attack. The repository also does not provide enough information to establish compatibility with Claude Desktop, Cursor, Windsurf, or Cline.

_Full upstream README: https://allmcps.com/mcp/zentric-protocol-mcp/readme_