Midas

local MCP server

Installation

Terminal

claude mcp add midas -e MIDAS_MCP_DB=... -e MIDAS_MCP_EMBEDDER=... -e MIDAS_MCP_MAX_RECORDS=... -e MIDAS_MCP_MIN_IMPORTANCE=... -- uvx midas-memory-mcp

Run in your project folder. Add --scope user to use it everywhere.

Claude Desktop, Cursor & other clients

Add to claude_desktop_config.json or .cursor/mcp.json.

{
  "mcpServers": {
    "midas": {
      "command": "uvx",
      "args": [
        "midas-memory-mcp"
      ],
      "env": {
        "MIDAS_MCP_DB": "...",
        "MIDAS_MCP_EMBEDDER": "...",
        "MIDAS_MCP_MAX_RECORDS": "...",
        "MIDAS_MCP_MIN_IMPORTANCE": "..."
      }
    }
  }
}

Fill in the environment variables with your own credentials.

Summary

Local-first, source-traceable agent memory — no LLM at ingest, fully offline

Questions

Midas is a local MCP server distributed as the pypi package midas-memory-mcp.

Run: claude mcp add midas -e MIDAS_MCP_DB=... -e MIDAS_MCP_EMBEDDER=... -e MIDAS_MCP_MAX_RECORDS=... -e MIDAS_MCP_MIN_IMPORTANCE=... -- uvx midas-memory-mcp

Run: codex mcp add midas --env MIDAS_MCP_DB=... --env MIDAS_MCP_EMBEDDER=... --env MIDAS_MCP_MAX_RECORDS=... --env MIDAS_MCP_MIN_IMPORTANCE=... -- uvx midas-memory-mcp

Similar MCP servers

NameStars

neurakeepio.github.dominiclachance/neurakeep11
agent-recallio.github.Goldentrii/agent-recall371
lean-memoryio.github.Wuesteon/lean-memory42
Cortex Memory Engineio.github.gambletan/cortex34
linksee-memoryio.github.michielinksee/linksee-memory14
Agent-Native Contentio.github.BuilderIO/agent-native-content6,874
Persomeio.github.Intuition-Lab/personal-model1,324
projectmemio.github.riponcm/projectmem844