saba-mcp

Model Context Protocol (MCP) server for Saba.

Saba is a persistent memory layer for AI coding agents. It stores engineering solutions, retrieves them semantically, and tracks evidence to verify what actually works.

This package exposes the three core MCP tools to connect any compatible AI assistant (like Claude Desktop or Cursor) to your Saba API server.

Tools Exposed

1. search_saba: Search collective engineering memory before solving a problem.
2. remember_saba: Save a new engineering discovery (project-specific behaviors, undocumented APIs, difficult bugs).
3. report_saba_outcome: Record whether a retrieved memory "worked" or "failed" to update the trust model.

Quick Start

1. Prerequisite

Ensure you have a running Saba API server. If you don't have one, see the Saba Repository to start the Fastify API server.

2. Connect your AI assistant via npx

Add saba-mcp directly to your MCP configuration file:

For Claude Desktop

Add to %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS/Linux):

{
  "mcpServers": {
    "saba": {
      "command": "npx",
      "args": ["-y", "saba-mcp"],
      "env": {
        "SABA_API_URL": "https://your-saba-api-url.com",
        "SABA_AGENT_ID": "your-developer-name"
      }
    }
  }
}

For Cursor / VS Code (with MCP Extensions)

Add a new stdio MCP server:

• Name: saba
• Command: npx -y saba-mcp
• Environment Variables:
  • SABA_API_URL: https://your-saba-api-url.com
  • SABA_AGENT_ID: your-developer-name

Configuration Variables

• SABA_API_URL: (Required) The URL where your Saba API server is hosted (e.g. http://localhost:3000).
• SABA_AGENT_ID: (Optional) Identifier for the agent or developer using the server (defaults to mcp-agent).