rememb — Claude Code MCP server: agent memory that survives sessions
Local memory for AI agents: project entries, keyword search and version history. No cloud, no API keys.
claude mcp add rememb -- uvx remembWhat rememb does
rememb keeps project context in a local store at ~/.rememb/ — entries.json by default, with meta.json and config.json beside it. An agent loads earlier decisions when a session starts, writes new ones as things change, and searches by keyword when it needs a narrow recall. Matching runs over entry text and tags; there are no embeddings and no calls to an outside service.
Entries live in sections: project, actions, systems, requests, user, context. Edits keep history, so you get diffs, restore of an earlier version and an audit trail. Limits, section list and storage backend sit in config.json; switch storage_backend to sqlite for bigger stores and existing JSON entries migrate on their own.
Install with pip install rememb; in MCP mode the store initializes on first use. A separate SSE transport (rememb mcp --transport sse --host 127.0.0.1 --port 8765) is for sharing one running process across several MCP clients on the same machine. Keep it out of a stdio client config — those clients expect JSON-RPC over stdin/stdout, while SSE serves an HTTP endpoint and starts on its own.
What it helps with
- stop re-explaining the stack and folder layout at the start of every session
- pull up an earlier decision by keyword instead of rereading everything
- roll a memory entry back to its previous version after a bad edit
- see what the agent wrote to memory and when
- keep a separate store per repository
Example prompts for the agent
- Save to rememb that the backend is PostgreSQL and auth lives in
src/auth/ - Search memory for anything about database migrations
- Show me which entries changed this session
How to install rememb
Platform CLI
claude mcp add rememb -- uvx remembuvx runs the command named after the package; if the README names another one, use uvx --from rememb <command>
Platform CLI
codex mcp add rememb -- uvx remembuvx runs the command named after the package; if the README names another one, use uvx --from rememb <command>
Config file
{
"mcpServers": {
"rememb": {
"command": "uvx",
"args": [
"rememb"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from rememb <command>
Config file
{
"servers": {
"rememb": {
"type": "stdio",
"command": "uvx",
"args": [
"rememb"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from rememb <command>
Platform CLI
gemini mcp add rememb uvx remembuvx runs the command named after the package; if the README names another one, use uvx --from rememb <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about rememb
How do I install rememb in Claude Code?
Run claude mcp add rememb -- uvx rememb, then restart your IDE. The agent can then read and write memory.
What does rememb do?
It adds a persistent memory layer for AI agents: structured project entries, keyword search, versioning and an audit trail. Everything stays in a local store.
Can several MCP clients share one rememb process?
Yes — start rememb mcp --transport sse --host 127.0.0.1 --port 8765 and connect through http://127.0.0.1:8765/sse. Do not put that mode into a stdio client config.
