mind-mem

mind-mem provides a drop-in memory layer that multiple MCP-compatible agents can share. It centralizes memory across Claude Code, Codex CLI, Gemini CLI, Cursor, Windsurf, Zed, OpenClaw, and any MCP-friendly client, enabling cross-agent recall, unified knowledge about projects and tools, and a governed audit trail. The server exposes a memory workspace and a set of governance-aware capabilities such as drift detection, contradiction handling, and structured persistence so agents can read and write to a single source of truth without fragmentation. Once running, agents connect to the shared memory workspace and begin to read from and write to the common memory store automatically, enabling cross-agent recall and consistent knowledge across tools.

To use mind-mem, first ensure the MCP server is running and accessible by your clients. Each MCP client will connect to the shared workspace, enabling unified memory across all supported agents. You can perform hybrid search and recall via the agents’ tooling, observe drift and contradictions across sessions, and leverage the Memory OS features to attach decisions, notes, and structured facts to the unified memory. The server is designed to be zero-infrastructure and locally hosted, so you can operate entirely on a developer workstation with a single shared workspace.

Prerequisites

• Python 3.10+ (recommended)
• pip (comes with Python)
• Git (optional, for cloning repositories)

Install the mind-mem MCP server

1. Install the Mind-Mem package which provides the memory backend and tooling:

pip install mind-mem

2. Install or prepare the MCP server component which exposes the memory workspace to MCP clients. If you are using the built-in server module, you can run it directly after installation:

# Run the MCP server (example)
python -m mind_mem_server

3. Initialize or configure a shared workspace for MCP clients:

# Create or initialize a workspace (if supported by your setup)
mind-mem-init ~/my-workspace

4. Verify connectivity from MCP clients by starting one or more supported agents and pointing them to the shared workspace. Typical usage involves setting an environment variable or configuration to specify the workspace path or endpoint where the memory is hosted.

Optional configuration

• Adjust the workspace path and database persistence as needed. If using on-disk persistence, specify a database path and ensure the server has read/write permissions.
• Review and tune governance options, drift detection sensitivity, and audit logging via environment variables or configuration files provided by mind-mem.

Troubleshooting tips

• Ensure Python 3.10+ is installed and available on your PATH.
• Confirm that the mind-mem server process is running and listening on the expected interface/port if applicable.
• Verify that MCP clients are configured to connect to the same shared workspace to enable cross-agent memory sharing.

Tips and notes:

• The mind-mem server is designed to be local-first and zero-infrastructure. Use a single workstation/workspace for best results.
• If you see drift or contradiction warnings, check the audit logs to trace the decision sources and update or repair the memory blocks accordingly.
• For multi-agent setups, ensure all clients reference the same workspace to enable true cross-agent memory sharing.
• If you need stronger persistence guarantees, consider configuring a durable on-disk database for the memory store and enabling WAL or equivalent transactional safeguards if supported by mind-mem.
• Review the MCP tooling docs for how to trigger recall, scan for drift, and apply governance rules across agents.