Recall

What is Recall?

Recall is an open-source MCP memory server that provides persistent, searchable, tiered memory for AI agents across sessions. It is designed for developers building multi‑session AI workflows with the Model Context Protocol.

How to use Recall?

Install via pip install ai-recallworks or run directly with uvx ai-recallworks stdio. For Cursor or Claude Desktop, add the provided JSON configuration calling uvx ai-recallworks stdio under mcpServers. Once running, use the built‑in tools (e.g., remember, recall) to store and retrieve memories.

Key features of Recall

• Persistent memory stored across AI agent sessions
• Semantic search over stored memories with recall
• Session lifecycle management via checkpoint and reflect
• Health monitoring and statistics with pulse
• Supports both stdio and HTTP+SSE transport modes
• MIT licensed with a separate commercial license available

Use cases of Recall

• Storing and retrieving agent notes across conversation sessions
• Enabling long‑term context retrieval for multi‑turn interactions
• Monitoring agent health and performance in production deployments
• Building remote, multi‑agent systems using HTTP+SSE mode

FAQ from Recall

How do I install Recall?

Use pip install ai-recallworks or run uvx ai-recallworks stdio directly (via uv).

What configuration is needed for Cursor or Claude Desktop?

Add a recall entry to the mcpServers object with command uvx and arguments ["ai-recallworks", "stdio"].

What tools does Recall provide?

It offers remember (store a note), recall (semantic search), checkpoint / reflect (session lifecycle), and pulse (health + stats).

Is there a remote deployment mode?

Yes, Recall supports HTTP+SSE mode for remote or multi‑agent setups.

What license does Recall use?

Recall is MIT licensed, with a separate commercial license available.