Context before conclusions.
Different environments don’t mean conflicting facts. Compare scope, roles, and version ranges to find real contradictions—not noise.
Context-aware analysis
The MCP server that finds contradictions across your code, docs, and deployments. Less guesswork. More ground truth.

Better context.
Wherever you build.
A consistency layer for your AI
Your stack has a lot to say. Make sure it's telling the same story.
Different environments don’t mean conflicting facts. Compare scope, roles, and version ranges to find real contradictions—not noise.
Context-aware analysis
Every finding points back to the exact source, line, or page. See what disagrees, why it matters, and which evidence holds more weight.
Source-level provenance
Get an actionable recommendation, not an uninvited change. Review, resolve, or dismiss—with an immutable trail of every decision.
Human-led resolution
Less noise. More signal.
See how context turns a simple comparison into a finding you can actually use.
# Production API service# Runtime environmentFROM node:22-alpineWORKDIR /appCOPY package*.json ./## Production requirements Node.js 18.x is required. Run npm install to get started.Both sources describe the production runtime, but Node.js 18.x and 22.x cannot be satisfied together. This is a real contradiction, not a formatting difference.
Real findings come with source authority, freshness scoring, and exact citations. No black boxes.
A little less fragmented
No new source of truth to maintain. Just a better understanding of the ones you already have.
Repositories, local documents, and public URLs. Bring in the places your team already works.
Extract claims, compare their context, and weigh each source by authority, freshness, and evidence.
Review a recommendation and choose what happens next. Every decision stays in your audit history.
Enterprise Verification Benchmark
How Contradiction MCP compares to naive vector search, LLM-as-a-judge prompting, and static syntax linters across 500+ multi-source assertions.
Local in-memory SQLite candidate generation
Deterministic SemVer algebra & leaf guards
Zero token burn; runs on local agent machine
Verifiable citations with immutable audit logs
| Evaluation Criterion | Contradiction MCPDedicated Engine | Vector RAG (Embeddings) | LLM-as-a-Judge Prompting | Static Rule Linters |
|---|---|---|---|---|
SemVer Satisfaction Calculus Computes intersecting SemVer ranges (^2.4.0 vs ~2.5.0) and isolates IP addresses from version math. | Deterministic SemVer algebra | Token string distance only | Prone to boundary hallucinations | No range algebra support |
Line-Level Evidence Provenance Pins every assertion to precise file paths, line numbers, and verbatim contextual snippets. | Exact file & line offset | Vague ~500-token chunk ranges | Approximate / fabricated quotes | Single-file syntax line only |
Heterogeneous Multi-Format Ingestion Correlates conflicting facts across Markdown, JSON, YAML, PDF, DOCX, Web pages, and GitHub APIs. | All 7 sources cross-analyzed | Text chunks (loses structure) | Constrained by context window | Single format specific |
False-Positive Suppression Namespace leaf guards prevent shared headings from conflating unrelated fields (e.g., host vs port). | Heading path + delimiter guards | High noise; similarity ≠ consistency | Biased by prompt phrasing | Rigid AST rules only |
Source Authority & Decay Scoring Weights evidence by origin hierarchy and applies exponential half-life decay to outdated docs. | AuthorityScorer + FreshnessScorer | Uniform embedding weights | Order & recency bias in prompt | No temporal awareness |
Stateful Human Review & Audit Trail Provides review lifecycle with chosen claim rationale and immutable append-only SQLite ledger. | Full lifecycle + audit ledger | Ephemeral retrieval only | Stateless responses | Process exit code only |
Air-Gapped Privacy & Security SSRF guard protects local network; zero data egress or third-party API exposure. | 100% Local / Air-Gapped | Vector DB egress risk | Transmits data to cloud APIs | Local execution |
Operational Cost per 10k Checks Total financial expenditure required to continuously verify assertions during CI/CD. | $0.00 (Zero API burn) | $8 – $25 (Vector compute) | $240 – $600 (Token pricing) | $0.00 |
Installation & Setup
Install with a single command into Claude, Cursor, Antigravity, or Windsurf. Published globally on npm. Zero configuration required.
Live on NPM · 1-Command Install · Local-First
# Configure in ALL detected IDEs automatically:npx -y contradiction-mcp install all # Or configure a specific editor:npx -y contradiction-mcp install antigravitynpx -y contradiction-mcp install cursornpx -y contradiction-mcp install claudenpx -y contradiction-mcp install claude-codenpx -y contradiction-mcp install windsurfWorks instantly on macOS, Windows, and Linux. No cloning or compilation needed.
Changelog & Release History
Latest updates, protocol revisions, and performance improvements shipped to Contradiction MCP.
Give your AI context it can stand behind.