Goldentrii/AgentRecall

Overview

AgentRecall is a persistent memory layer for AI agents that works across sessions and platforms via the Model Context Protocol (MCP). It stores corrections and learnings from agent mistakes, compresses context to reduce token usage, and consolidates knowledge overnight. The tool is available as an npm package (agent-recall-mcp) and is written in TypeScript.

Best for

Best for
Developers running multiple AI agent sessions who want persistent, cost-efficient memory that improves from errors.

Use cases

• Give an agent long-term memory across Claude Code, Codex, and Gemini sessions
• Automatically capture and apply corrections from agent mistakes
• Reduce token costs by compressing and consolidating context overnight

Notes

AgentRecall is a persistent memory layer for AI agents that works across sessions and platforms via the Model Context Protocol (MCP). It stores corrections and learnings from agent mistakes, compresses context to reduce token usage, and consolidates knowledge overnight. The tool is available as an npm package (agent-recall-mcp) and is written in TypeScript.

303 stars on GitHub. Last updated 2026-06-18. Licensed MIT.

Use cases

• Give an agent long-term memory across Claude Code, Codex, and Gemini sessions
• Automatically capture and apply corrections from agent mistakes
• Reduce token costs by compressing and consolidating context overnight

Pros

• Cross-platform support via MCP works with multiple agent tools
• Reduces token usage through automatic context compression
• Learns from mistakes without manual intervention

Cons

• Requires MCP-compatible client to function
• Overnight consolidation may delay knowledge updates
• Limited to correction-driven memory, not general-purpose storage

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.