SLM MCP Hub
The World's First MCP Gateway That Learns
One hub process. Every MCP server. Every AI client. SLM MCP Hub federates 400+ tools through 3 meta-tools, recovers 150K tokens of context per session, and shares cache and cost telemetry across Claude Code, Cursor, Windsurf, and VS Code. Pair it with SuperLocalMemory and the hub learns from every tool call.
The Context Window Tax
Every AI coding session spawns its own MCP server processes. Five Claude Code sessions with 36 MCPs means 180 OS processes eating ~9 GB of RAM. Every session also loads ~150K tokens of tool definitions into the context window before you type a word. On a 200K-context model, 75% is gone before the agent thinks. And every session starts from zero — no shared cache, no cost tracking, no learning, no coordination.
Key Capabilities
3 Meta-Tools, 430+ Tools
Instead of loading 400+ tool schemas into every session, clients get 3 meta-tools: hub__search_tools, hub__call_tool, hub__list_servers. The hub routes to the right server on demand.
79% Process Reduction
One hub process for all MCP servers. 5 sessions × 36 MCPs collapses from 180 processes (~9 GB RAM) to 37 processes (~1.9 GB RAM). Sessions connect over HTTP instead of spawning 36 child processes each.
150K Tokens Saved Per Session
Tool schemas are discovered on demand, not preloaded. Recover 75% of a 200K context window that was being consumed by tool definitions before the agent even started.
Federated Cache + Cost Intelligence
Every connected client shares the same cache, cost ledger, and trace log. Pair the hub with SuperLocalMemory and it learns which tool calls to cache, skip, or redirect based on past outcomes.
One Config, Every IDE
Configure MCPs once in the hub. Claude Code, Cursor, Windsurf, VS Code, and any MCP-compatible client share the same tool set via a single HTTP endpoint. No per-IDE config duplication.
SuperLocalMemory
Information-Geometric Memory for AI Agents
Local-first AI agent memory with mathematical foundations. 74.8% on LoCoMo without cloud dependency — highest local-first score reported. Fisher-Rao retrieval, sheaf cohomology, Langevin lifecycle. EU AI Act compliant.
SLM Mesh
Peer-to-peer communication layer for AI coding agents
P2P agent communication with 8 MCP tools. Agents discover each other, send messages, share state, and lock files. Works with any MCP-compatible agent. SQLite + UDS for <100ms delivery. 480 tests, 100% coverage.
AgentAssert
Design-by-Contract for AI Agents
Formal specification and runtime enforcement of behavioral contracts for autonomous AI agents. Prevents drift, ensures compliance, enables composition.