Developer Tooling

Model Context Protocol (MCP) Server Integration

MBHub includes a built-in, native Model Context Protocol (MCP) daemon running over standard I/O (stdio). You can hook Cursor, Claude Desktop, VS Code, and Windsurf directly into MBHub so coding agents query your local and collective memory before spending cloud tokens.

1. The Frontier Augmentation Concept

Developer tools and local agents often run smaller models (3B to 14B parameters) due to latency and cost. While fast, these models frequently struggle with complex, obscure, or bleeding-edge API questions.

Augment Local Agents with Collective Wisdom

By connecting your agent to MBHub via MCP, when your agent needs to know “How to implement distributed consensus in Rust with Raft?”, it queries MBHub’s L1/L2 memory first. If any peer worldwide has already resolved this inquiry with a frontier model (Claude 3.7 Sonnet, GPT-4o, DeepSeek-V3), the verified answer is injected into your agent’s context at sub-5ms latency for $0.00.

2. Configuration Files

Claude Desktop Configuration

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) %APPDATA%\Claude\claude_desktop_config.json (Windows)

{
  "mcpServers": {
    "mbhub": {
      "command": "mbhub",
      "args": ["mcp"]
    }
  }
}

Cursor MCP Configuration

Settings > Features > MCP > Add New MCP Server

Name: mbhub
Type: command
Command: mbhub mcp

Windsurf & VS Code Settings

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "mbhub": {
      "command": "mbhub",
      "args": ["mcp"]
    }
  }
}

3. MCP Tools Exposed by MBHub

Tool Name	Description	Parameters
`mbhub_search`	Searches local L1 SQLite and L2 P2P swarm for matching atomic inquiries.	`query` (string, max 80 chars), `threshold` (optional float, default 0.85)
`mbhub_get_by_hash`	Retrieves verified answer by exact BLAKE3 content hash.	`hash` (string, 64-char hex)
`mbhub_status`	Returns local node health, connected peer count, and cache statistics.	None