MCP Registry Manager

MCP Registry Manager 🌐

Centralized discovery and quality scoring for the exploding MCP (Model Context Protocol) ecosystem.

What It Does

The MCP ecosystem is growing fast — awesome-mcp-servers, AllInOneMCP, GitHub — but no unified discovery or quality checks.

MCP Registry Manager provides:

• Unified discovery — Aggregate servers from multiple sources
• Quality scoring — Test coverage, documentation, maintenance status
• Semantic search — "Find servers for file operations" (not just keyword search)
• Install management — Install/uninstall with dependency resolution
• Categorization — Organize by domain (files, databases, APIs, dev tools)

Problem It Solves

MCP is becoming the "USB-C of agent tools" but:

• Discovery is fragmented (GitHub repos, lists, registries)
• No quality signals (which servers are production-ready?)
• No semantic search (can't find "what does this do?")
• No unified management

Usage

# Discover all MCP servers from GitHub
python3 scripts/mcp-registry.py --discover

# Discover local MCP servers from config files
python3 scripts/mcp-registry.py --discover-local

# Search semantically
python3 scripts/mcp-registry.py --search "file system operations"

# Get quality report for a server
python3 scripts/mcp-registry.py --score @modelcontext/official-filesystem

# Install a server
python3 scripts/mcp-registry.py --install @modelcontext/official-filesystem

# List installed servers
python3 scripts/mcp-registry.py --list

# List only installed servers
python3 scripts/mcp-registry.py --list --installed

# Check health of a specific server
python3 scripts/mcp-registry.py --health-check local/filesystem

# Check health of all installed servers
python3 scripts/mcp-registry.py --health-all

# Export registry to JSON
python3 scripts/mcp-registry.py --export registry.json

Quality Score Formula

Quality = (0.4 * TestCoverage) + (0.3 * Documentation) + (0.2 * Maintenance) + (0.1 * Community)

Where:
- TestCoverage = % of code covered by tests
- Documentation = README completeness, API docs, examples
- Maintenance = Recent commits, responsive issues
- Community = Stars, forks, contributors

Data Sources

Local Discovery Scans These Paths

The --discover-local flag scans for MCP server configs in these locations:

• ~/.config/claude/mcp_servers.json
• ~/.config/mcp/servers.json
• ~/.openclaw/config/mcp_servers.json
• ~/.claude/mcp_servers.json

Local servers are automatically:

• Tagged with category local
• Given maximum quality score (1.0)
• Stored with their config path for reference

Categories

• Files — Filesystem, storage, S3
• Databases — PostgreSQL, MongoDB, Redis, SQLite
• APIs — HTTP, GraphQL, REST
• Dev Tools — Git, Docker, CI/CD
• Media — Image processing, video, audio
• Communication — Email, Slack, Discord
• Utilities — Time, crypto, encryption

Architecture

┌─────────────────┐
│  Discovery      │  ← awesome-mcp, GitHub, AllInOneMCP
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Registry DB    │  ← SQLite/PostgreSQL with metadata
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Quality Scorer │  ← Test coverage, docs, maintenance
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Semantic Search│  ← Embeddings + vector search
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  CLI Interface  │  ← Install/uninstall/update
└─────────────────┘

Requirements

• Python 3.9+
• requests (for GitHub API)
• sentence-transformers (for semantic search)
• numpy/pandas (for scoring)

Installation

# Clone repo
git clone https://github.com/orosha-ai/mcp-registry-manager

# Install dependencies
pip install requests sentence-transformers numpy pandas

# Run discovery
python3 scripts/mcp-registry.py --discover

# Discover local servers
python3 scripts/mcp-registry.py --discover-local

Health Checks

Monitor the status of your installed MCP servers with built-in health checks:

Single server:

python3 scripts/mcp-registry.py --health-check local/filesystem

All installed servers:

python3 scripts/mcp-registry.py --health-all

Health checks include:

• url_valid — URL exists and is properly formatted
• url_type — Transport type (stdio, sse, http/ws)
• config_valid — Has required config fields (command or url)
• has_description — Server has a description

Status levels:

• healthy — All checks pass
• degraded — Some checks pass, config is valid
• unhealthy — Critical checks fail

Inspiration

• MCP Server Stack guide — Essential servers list
• awesome-mcp-servers — Community-curated directory
• AllInOneMCP — Remote MCP registry
• Klavis AI — MCP integration platform

Local-Only Promise

• Registry metadata is cached locally
• Install operations run locally
• No telemetry or data sent to external services

Version History

• v0.2 — Local MCP discovery + health checks
• v0.1 — MVP: Discovery, quality scoring, semantic search
• Roadmap: GitHub integration, CI tests, auto-updates