AI & Memory MCP Server

Core

Name: Core
Author: omega-memory

Persistent memory for AI coding agents. #1 on LongMemEval. Local-first.

Scan found nothing by omega-memory41Apache-2.0PythonIntermediatestdio

Install

Claude Code

claude mcp add core -- uvx core

Cursor

{
  "mcpServers": {
    "core": {
      "command": "uvx",
      "args": [
        "core"
      ]
    }
  }
}

VS Code

{
  "mcp": {
    "servers": {
      "core": {
        "command": "uvx",
        "args": [
          "core"
        ]
      }
    }
  }
}

Windsurf

{
  "mcpServers": {
    "core": {
      "command": "uvx",
      "args": [
        "core"
      ]
    }
  }
}

Claude Desktop

{
  "mcpServers": {
    "core": {
      "command": "uvx",
      "args": [
        "core"
      ]
    }
  }
}

Safety Report

43% Scan found nothing

Scanned 5 months ago

3 passed 4 warnings

Security

Authentication

details

Authentication detected: env_required, bearer_check, oauth

Pass

CORS Policy

details

stdio transport — CORS not applicable

N/A

Rate Limiting

details

Rate limiting detected: rate_limit_middleware, throttle

Pass

Known CVEs

details

1 known vulnerable deps (1 high)

Warning

Code Quality

Dependency Audit

details

1 known vulnerable deps (1 high)

Warning

Dangerous Patterns

details

Dangerous patterns detected: eval_call

Warning

License

details

License: Apache-2.0

Pass

Community

Maintenance Status

details

Last commit 5 months ago

Warning

Learn about our security methodology →

Compatibility

Claude CodeFull support

Tested

CursorFull support

Tested

VS CodeFull support via Copilot

Tested

WindsurfFull support

Tested

Claude DesktopFull support

Tested

Frequently Asked Questions

What is Core?

Persistent memory for AI coding agents. #1 on LongMemEval. Local-first.

Is Core safe to use?

We cannot tell you that, and nobody scanning at this scale can. What we did in March 2026 was shallow-clone Core and run five keyword searches over its source: auth identifiers, CORS configuration, rate-limit identifiers, the dependency manifest against a hardcoded list of 18 known-vulnerable packages, and unguarded eval/exec calls. None of them matched. A text search cannot follow a variable, judge intent, or tell you what a server does once it is running, so "nothing matched" is not "safe". What the scan does and does not cover is written out at /mcp/security.

What are alternatives to Core?

Similar MCP servers include Context7, Cognee, Hindsight. Each serves a similar purpose but may differ in features, language, and compatibility.