# Daksh-create349/Contradiction-MCP [Health: Active]

**Category:** 📐 Architecture & Design  
**Repository:** https://github.com/Daksh-create349/Contradiction-MCP  
**GitHub Stars:** 5  
**npm Downloads (last month):** 974  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/daksh-create349-contradiction-mcp

## Description
Detect contradictions and architectural drift in Kubernetes manifests, READMEs, OpenAPI specs and Word docs, with deterministic scoring and resolution advice.

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

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

## Documentation

## What Daksh-create349/Contradiction-MCP MCP server does

Daksh-create349/Contradiction-MCP MCP server audits multiple sources of engineering information for conflicting claims and architectural drift. It can inspect Kubernetes and other YAML content, JSON, Markdown, CSV, plain text, RFC822 email files, OpenXML Word documents, repository files, and public web pages. Its GitHub connector covers runtime engines, Dockerfiles, GitHub Actions workflows, and READMEs. The document connector provides page- and line-based evidence for supported document formats.

The server is intended for the Developer Tools category and fits workflows where configuration, deployment files, API descriptions, and written guidance need to agree. Examples include checking whether documented runtime versions match deployment manifests, finding inconsistent ports or URLs, and comparing environment-specific claims without treating production and staging as equivalent.

## How it works

Daksh-create349/Contradiction-MCP MCP server extracts explicit claims and compares them according to semantic context. Context can include environment, scope, and source role, such as production versus staging, file versus cluster, or documentation versus deployment. Version comparisons use semantic-version range logic rather than simple string matching. The implementation also distinguishes IPv4, IPv6, host, and port values so unrelated value types are not compared as versions.

Markdown tables, headings, code fences, bullet points, and Shields.io or GitHub Actions status badges are parsed for relevant claims. Namespace-aware leaf matching reduces false matches between unrelated properties under shared headings, while matrix and delimiter parsing handles multi-column tables.

Conflicting claims receive authority, freshness, and evidence assessments. Authority considers source hierarchy and origin credibility; freshness applies time-based decay; and evidence evaluation considers citation directness and snippet quality. A resolution advisor produces recommendations without changing state automatically. Operators can move findings through open, reviewed, resolved, dismissed, and reopened states, with append-only audit history.

## Setup and configuration

The project requires Node.js 20 or newer and uses TypeScript. It supports standard MCP over Stdio and Streamable HTTP/SSE. The Stdio mode is documented for Claude Desktop and Cursor. HTTP/SSE mode uses Bearer API-key authentication and sliding rate limiting.

File ingestion is constrained by configured allowed roots, defaulting to the current workspace. The website connector applies SSRF protections and limits redirects to five hops. These boundaries should be considered when deciding where the server runs and which sources it can inspect.

## Tools and capabilities

The MCP interface is described as providing 12 canonical tools, along with four resources and two prompts. The supplied material does not list each tool name, but identifies capabilities for source ingestion, claim creation, contradiction analysis, scoring, advisory resolution, and lifecycle review. The `create_claim` operation can be used when heuristic extraction is insufficient and a verified predicate needs to be supplied programmatically.

The embedded SQLite store uses Write-Ahead Logging, while filesystem access includes directory-containment checks. Website retrieval performs DNS validation and blocks loopback, private network, and cloud metadata destinations. These controls support audits that combine local files, repository content, and externally hosted documentation.

## Limitations and notes

Deeply nested JSON, YAML, Kubernetes, or Terraform-like structures can produce false positives when identical leaf keys occur in separate blocks. Full JSON-path namespace scoping is not yet available, so cross-document comparisons are safer than unrestricted intra-file comparisons.

Extraction is heuristic for prose. Nuanced negation, implied conditions, and statements that describe rejected alternatives may not be interpreted correctly. Structured key-value declarations or explicit `create_claim` inputs are better for assertions that require precise predicate semantics.

The website connector does not execute client-side JavaScript. Single-page applications must provide static HTML snapshots or rendered Markdown. Files outside allowed roots and directory traversal attempts are rejected, and private, loopback, and metadata-network destinations are blocked.

_Full upstream README: https://allmcps.com/mcp/daksh-create349-contradiction-mcp/readme_