Skip to content

MCP_SETUP.md

Latest commit

 

History

History
102 lines (76 loc) · 2.88 KB

MCP_SETUP.md

File metadata and controls

102 lines (76 loc) · 2.88 KB

SAFLA MCP (Model Context Protocol) Setup

Overview

SAFLA is now configured to work with the Model Context Protocol (MCP), allowing AI assistants to interact with the remote SAFLA instance deployed on Fly.io.

Configuration

The MCP configuration is stored in .roo/mcp.json:

{
  "mcpServers": {
    "safla": {
      "command": "python3",
      "args": [
        "/workspaces/SAFLA/safla_mcp_simple.py"
      ],
      "env": {
        "SAFLA_REMOTE_URL": "https://safla.fly.dev"
      }
    }
  }
}

Available Tools

The SAFLA MCP server provides the following tools:

1. generate_embeddings

Generate embeddings using SAFLA's extreme-optimized engine (1.75M+ ops/sec)

  • Input: texts (array of strings) - Texts to embed
  • Performance: Utilizes the 178,146% optimized engine

2. store_memory

Store information in SAFLA's hybrid memory system

  • Input:
    • content (string) - Content to store
    • memory_type (string) - Type: "episodic", "semantic", or "procedural"

3. retrieve_memories

Search and retrieve from SAFLA's memory system

  • Input:
    • query (string) - Search query
    • limit (integer, default: 5) - Maximum results

4. get_performance

Get SAFLA performance metrics

  • Input: None required
  • Output: Current performance statistics

MCP Server Implementation

The MCP server (safla_mcp_simple.py) implements the Model Context Protocol by:

  1. Reading JSON-RPC messages from stdin
  2. Proxying requests to the remote SAFLA API
  3. Returning formatted responses to stdout

Performance Characteristics

  • Baseline: 985.38 ops/sec
  • Optimized: 1,755,595.48 ops/sec
  • Improvement: 178,146.95% (1,781x faster)
  • Cache: Enabled for maximum performance
  • Batch Size: 256 (optimal for performance)

Testing

To test the MCP server manually:

# Start the server
python3 /workspaces/SAFLA/safla_mcp_simple.py

# Send test requests (JSON-RPC format)
{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {}}
{"jsonrpc": "2.0", "id": 2, "method": "tools/list", "params": {}}

Remote SAFLA Instance

The MCP server connects to the SAFLA instance deployed at:

Troubleshooting

If you encounter connection issues:

  1. Verify the SAFLA instance is running: curl https://safla.fly.dev/health
  2. Check the MCP server logs for errors
  3. Ensure the SAFLA_REMOTE_URL environment variable is set correctly
  4. Verify Python 3 is available in your environment

Architecture

AI Assistant <-> MCP Protocol <-> safla_mcp_simple.py <-> HTTPS <-> SAFLA on Fly.io

The MCP server acts as a bridge between the Model Context Protocol and the remote SAFLA API, enabling seamless integration with AI assistants while leveraging the extreme performance optimizations achieved (1.75M+ ops/sec).