MCP Server Guide — Model Context Protocol for AI Coding

What Is MCP?

Model Context Protocol (MCP) is an open standard that lets AI tools access external data sources and tools. Think of it as a bridge between your AI assistant and your local environment — files, databases, APIs, and memory engines.

Before MCP, you had to paste files manually into your AI chat. With MCP, your AI can read files, query databases, and access memory engines directly.

How MCP Works

MCP uses a client-server architecture:

• MCP Client — The AI tool (Claude Desktop, Cursor, Continue) that connects to MCP servers
• MCP Server — A local process that exposes tools and resources to the client
• Tools — Functions the AI can call (search files, query memory, read documents)
• Resources — Data the AI can access (file contents, database records, memory chunks)

MCP-Compatible Tools

Setting Up an MCP Memory Server

The most powerful use of MCP is giving your AI persistent memory. Here's how to set up Eidos Memory as an MCP server:

For Claude Desktop

# Generate config
eidos mcp print-config --client claude-desktop --copy

# Or manually add to claude_desktop_config.json:
{
  "mcpServers": {
    "eidos": {
      "command": "eidos",
      "args": ["mcp", "serve"]
    }
  }
}

For Cursor

# Generate config
eidos mcp print-config --client cursor --copy

# Or manually add to .cursor/mcp.json:
{
  "mcpServers": {
    "eidos": {
      "command": "eidos",
      "args": ["mcp", "serve"]
    }
  }
}

MCP Tools Available in Eidos

Eidos provides 14 MCP tools:

• eidos_search — Semantic search across your codebase
• eidos_assemble — Build context for a specific query
• eidos_remember — Store a decision or fact
• eidos_recall — Retrieve stored decisions
• eidos_feedback — Rate context quality for learning
• eidos_graph — Query the knowledge graph
• eidos_status — Check memory engine status
• eidos_index — Index a directory
• eidos_projects — List indexed projects
• eidos_chunks — List context chunks
• eidos_sessions — Manage sessions
• eidos_config — View/update configuration
• eidos_export — Export memory data
• eidos_import — Import memory data

The Bottom Line

MCP is the standard that makes AI coding tools extensible. A memory server like Eidos is the most impactful MCP server you can add — it gives every AI tool persistent memory, context injection, and 95% token savings.