Hey all — sharing something I’ve been running locally for a few months and just open-sourced.
Recall is an MCP server that gives agents persistent, searchable memory across sessions. Drop it into Cursor’s MCP config and your agent stops forgetting yesterday’s context.
- pip install ai-recallworks (or run via uvx)
- Works over stdio (Cursor / Claude Desktop) or HTTP+SSE (remote agents)
- Per-agent namespaces so multi-agent setups don’t cross-contaminate
- Tiered memory: session / durable / reference / anti-pattern
- Tools: remember, recall, recall_filtered, checkpoint, reflect, pulse, session_close, anti_pattern, reindex, forget, index_file, memory_stats
- MIT licensed (separate commercial license available)
Site: https://www.recall.works
Repo: GitHub - RecallWorks/Recall: Open-source memory for AI agents. MCP-native. Self-hosted. One Docker image. · GitHub
PyPI: search “ai-recallworks”
Cursor config snippet:
{
"mcpServers": {
"recall": {
"command": "uvx",
"args": ["ai-recallworks", "stdio"]
}
}
}
Feedback / issues / PRs welcome.