Lightning MCP

What is MCP?

What Lightning MCP Is

Lightning MCP is an enterprise aggregation and persistence layer built on top of MCP. We don’t replace MCP — we extend it with everything production AI infrastructure needs.

Architecture

Everything flows through a single path: your AI client connects to the Bridge, the Bridge routes to the Singularity Engine, and the Engine dispatches to the right module.

Singularity is modular. Adding a new integration means dropping in a Python file — no restart, no reconfiguration. The architecture supports any tool that can expose an API.

What’s Live Today

36 modules running in production across 3 nodes. Every module is independently loadable.

What’s Next

The modular architecture means any API-accessible service can become a Singularity module. These are on the roadmap or ready to build when a customer needs them:

Single Endpoint

Why This Exists

When Anthropic released MCP, we were excited. We immediately started connecting Claude to everything — our servers, Docker, Proxmox, databases, firewall, Home Assistant, phone system. Each one was its own MCP server with its own connection.

By the time we had 15 connections, Claude’s performance had visibly degraded. Response times went up. The AI spent more time managing tool connections than actually thinking about our questions. Context windows filled up with tool definitions instead of useful conversation. It was unusable.

That’s why we built Lightning MCP. One endpoint that aggregates everything. Claude connects once, sees all 180+ tools, and performs like it only has one connection — because it does.

mcp_servers:
  - filesystem:     localhost:8821
  - postgres:       localhost:8823
  - redis:          localhost:8824
  - docker:         localhost:8825
  - proxmox:        localhost:8826
  - home-assistant: localhost:8828
  - pfsense:        localhost:8829
  - voipms:         localhost:8830
  ... # 15+ more connections
  
# Each adds ~200 tokens of tool defs
# Context fills up fast
# Response time degrades with every server
# One server crashes = broken session

{
  "mcpServers": {
    "lightning": {
      "url": "https://lmcp.your.ai/mcp",
      "transport": "sse"
    }
  }
}

# 1 connection
# 180+ tools
# Internal routing, zero overhead
# Bridge handles failover automatically
# Any module can hot-reload
# No context bloat

Works with Claude Desktop, Claude Code, Cursor, Windsurf, Cline, and any client that speaks MCP over SSE.

Persistent Memory

Standard MCP has no memory between sessions. Lightning adds a 3-tier persistence architecture so your AI never starts from zero.

Full breakdown of the memory architecture →

Active-Active Failover

Not active-passive. All three nodes accept requests simultaneously. KeyDB handles replication and conflict resolution. If any node goes down, the others are already running — zero switchover latency.

Explore Further