xone-mcp-risk-index 0.3.1

Evidence-backed MCP server risk signal catalog and local CLI.

mcp-risk-index

Languages: English | 中文

An open risk index for common MCP servers, permissions, commands, and maintenance signals.

Status

v0.3.1 - evidence-backed catalog rendering with reviewer questions, strict review checks, and refreshed public sample metadata.

Purpose

Convert mcp-audit rule experience into a reusable public data asset without unsupported claims.

First Production Surface

Versioned data catalog with evidence-backed entries and a deterministic local CLI.

From PyPI:

python3 -m pip install xone-mcp-risk-index
mcp-risk-index init --output mcp-risk-index.catalog.yml
mcp-risk-index validate --catalog mcp-risk-index.catalog.yml --strict
mcp-risk-index render --catalog mcp-risk-index.catalog.yml --format markdown --output mcp-risk-index.md
mcp-risk-index render --catalog mcp-risk-index.catalog.yml --format json --output mcp-risk-index.json

From Homebrew:

brew install x-one-ai/tap/mcp-risk-index
mcp-risk-index --version

From a source checkout, you can also validate the bundled catalog:

mcp-risk-index validate --catalog data/catalog.yml --strict
mcp-risk-index render --catalog data/catalog.yml --format markdown --output mcp-risk-index.md
mcp-risk-index render --catalog data/catalog.yml --format json --output mcp-risk-index.json

For local development:

python3 -m pip install -e '.[dev]'
python3 -m pytest tests -q

Catalog Contract

The bundled catalog uses mcp-risk-index.catalog.v1. Each entry records identity, package, launch command, permissions, maintenance facts, review-level risk signals, evidence, and limitations. Rendered Markdown and JSON also include reviewer questions derived from the evidence-backed signals.

Review levels are prompts for human inspection:

• info: useful context
• review: inspect before adoption
• high-review: require explicit owner approval

They are not safety scores.

Strict validation requires production review governance fields such as maintenance.source_checked_at and a GitHub repository source.

Required Evidence

• server identity
• permission profile
• command/package signals
• maintenance signals
• evidence links

Non-Goals

• no subjective ranking without evidence
• no broad repo health clone
• no security claims without criteria
• no absolute safe/unsafe labels

OPT Operating Model

This project references the shared One Person Team workflow through ops/opt-overlay.md. Project-specific constraints live under ops/constraints, and evolvable local skills live under ops/skills.

Blocked Inputs

Inputs that require user or real-world data are recorded in ../x-one-skipped-inputs.md and should not block foundation work.

Real-user feedback should be classified as false-positive, false-negative, adapter-request, scenario-request, or catalog-update when it applies; portfolio-level handling is tracked in X-One portfolio health docs.

Docs

• Product Foundation
• Catalog Governance
• MCP Audit Adoption Review
• Risk Context Note
• Publishing
• Homebrew Packaging
• Catalog Design
• OPT Overlay
• Production Constraints
• Main Entry Constraints
• Skill Evolution