“Can you explain the repository, architecture and previous decisions again?”
Open-source Model Context Protocol server
Your AI coding tools forget. This remembers.
Persistent Memory MCP stores architecture, decisions, tasks, warnings, file relationships and session state so Codex, Claude Code, OpenCode, Qwen Code and other MCP clients can continue the same project without losing context.
Resume the same project in another AI client
$ memory-mcp doctor ✓ Python 3.12 ✓ Supabase connected ✓ OWNER_ID configured You: Resume this project. AI: Authentication refactor is 70% complete. Active warning: review RLS policies. Next step: implement token rotation.
“I loaded the latest checkpoint, three open tasks, one active warning and the files involved in the refactor.”
One durable memory layer for every AI coding agent
Keep project history structured, searchable and portable instead of burying it inside one chat or one vendor.
Cross-client continuity
Move between Codex, Claude Code, OpenCode and other MCP clients while preserving the same project state.
Git-aware context
Remember repository, remote, branch, commit and working-tree status alongside project memory.
Structured decisions
Store why technical choices were made, not only what files changed.
Tasks and checkpoints
Resume from the exact implementation state with blockers and next actions.
Searchable memory
Use semantic retrieval with a lexical fallback to find only the context that matters.
Privacy controls
Use Supabase Row Level Security, exports and retention policies to control stored context.
Install and configure in minutes
The distribution is named persistent-memory-mcp. Both memory-mcp and persistent-memory-mcp commands are available.
Install the package
pipx install persistent-memory-mcp
Create configuration interactively
memory-mcp init
This writes a private .env file, generates an MCP configuration block and tests Supabase connectivity.
Install the database schema
Open the Supabase SQL Editor and run the repository's schema.sql file once.
Check the installation
memory-mcp doctor memory-mcp status
Standard MCP configuration
Use the same server entry in any client that supports mcpServers.
{
"mcpServers": {
"persistent-memory-mcp": {
"command": "memory-mcp",
"env": {
"SUPABASE_URL": "https://your-project.supabase.co",
"SUPABASE_KEY": "your-anon-key",
"OWNER_ID": "your-stable-identifier"
}
}
}
}Codex
Load shared project context before implementation and preserve important decisions at the end of a session.
Claude Code
Continue architecture and refactoring work without reconstructing previous conversations.
OpenCode and Qwen Code
Reuse the same memory backend and standard MCP command configuration.
Prompts that work naturally
“Resume this project and tell me where we left off.”
“Save the architecture decision we just made.”
“Show active warnings before changing authentication.”
“Remember the important files modified in this session.”
“Search memory for the database migration decision.”
“Save everything important from this session.”
Frequently asked questions
What is Persistent Memory MCP?
It is a Python MCP server that stores structured software-project memory so AI agents can continue work across sessions and clients.
Which tools are supported?
Codex, Claude Code, Claude Desktop, OpenCode, Qwen Code and other clients that support standard Model Context Protocol server configuration.
Where is memory stored?
The current backend uses Supabase and PostgreSQL. The schema includes projects, sessions, decisions, tasks, warnings, checkpoints, file memory, timeline events and searchable memory documents.
Does it replace Git?
No. Git stores code history. Persistent Memory MCP stores the project reasoning, implementation state and context that code history does not explain by itself.
Is it open source?
Yes. The project uses the MIT License and welcomes contributions.