A technical overview of what we built on top of Anthropic’s Model Context Protocol, why it matters, and how the pieces fit together.
MCP is an open standard from Anthropic that lets AI models connect to external tools and data sources. Instead of building custom integrations, MCP provides one standardized way for AI to interact with the outside world.
It defines how an AI client discovers tools, calls them, and receives results — with built-in type safety and error handling.
In standard MCP, you run individual servers — one for your filesystem, one for your database, one for Docker. Each needs its own config and auth. At scale you’re managing dozens of separate connections.
# Standard MCP: one server per tool mcp_servers: - filesystem: localhost:8821 - postgres: localhost:8823 - redis: localhost:8824 - docker: localhost:8825 - home-assistant: localhost:8828 ... # dozens more to maintain
We don’t replace MCP. We extend it with everything production AI infrastructure needs.
# Lightning MCP: one endpoint, everything mcp_servers: - lightning: https://lmcp.yourdomain.com/mcp
180+ tools from 23 modules through a single endpoint. Auto-discovery, unified auth, zero per-tool configuration.
16+ per-project Redis databases with 3-way active-active replication across nodes. AI that remembers across sessions. Deep dive →
Unimatrix vector database with 4,264 embeddings. Search all projects by meaning, not keywords.
3 independent nodes all serving simultaneously. Zero switchover latency — no single point of failure.
Per-project operational awareness that survives across sessions. The AI knows what it should do every time it starts.
Built-in ITSM with incidents, changes, projects, tasks, and 500+ knowledge articles. Enterprise audit trails out of the box.
Lightning MCP runs on a 3-node distributed architecture. Every node can serve requests independently.
Primary dev + production. All MCP tools, Unimatrix, ITSM, Daily Planet.
Hot failover. Full KeyDB replica. Independent Singularity instance.
Third replica. PostgreSQL backup. ZFS archive. Full data redundancy.
| Component | Purpose |
|---|---|
| Lightning Bridge | MCP aggregation proxy — single endpoint for all tools |
| Singularity Engine | Tool execution — routes calls to the right module |
| Unimatrix (Qdrant) | Vector database — semantic search across all memories |
| ITSM API | Service management — incidents, changes, knowledge base |
| Lightning Forge | Code validation — syntax, security, best practices |
| Daily Planet | Internal monitoring — AI newsroom with autonomous reporters |
| Continuum | Conversation archival — streams live chats to storage |
// Claude Desktop
{
"mcpServers": {
"lightning": {
"url": "https://lmcp.yourdomain.com/mcp",
"transport": "sse"
}
}
}
// Works with Claude Desktop, Claude Code,
// Cursor, Windsurf, Cline, or any MCP client
"Check Docker containers" → docker_ps "What was I working on yesterday?" → recall (memory search) "Turn off the living room lights" → ha_call (Home Assistant) "Validate this Python code" → forge_validate "Create an incident for the outage"→ itsm_create
Deep dive into the 3-tier memory architecture and Unimatrix vector search.
Full platform overview with tool modules and endpoint details.
Learn about enterprise deployments and custom configurations.