An MCP server that gives AI agents a shared, persistent memory graph across sessions and agent boundaries

Every AI agent session starts from zero. Context, preferences, and prior decisions vanish the moment a session ends. mem0 hit 58.5k GitHub stars and raised $24M Series A solving this for single-agent flows, but the real gap is bigger: when two agents (a coding agent and a project-management agent, or two Claude Code sessions) work on the same project, they each maintain isolated silos with no shared knowledge. A cross-agent, cross-session memory MCP server solves this by maintaining a structured knowledge graph that any MCP-compatible agent can read and write. Developers connect their agents once via the MCP protocol; from that point every agent sees prior decisions, user preferences, project context, and inter-agent handoffs as a queryable graph. The SaaS backend handles sync, deduplication, and access control across agents and team members.

Demand Breakdown

GitHub

65,200

262

Social Proof 2 sources

mem0ai/mem0: AI agents forget everything between sessions, statelessness breaks multi-turn workflows

@deshraj · 2024-09-01

65,200 HN

Show HN: Mem0 - open-source Memory Layer for AI apps

@staranjeet · 2024-09-06

262

Gap Assessment

CompetitiveMultiple tools exist but differentiation opportunities remain

4 tools exist (mem0, Zep, Letta (formerly MemGPT), Supermemory) but gaps remain: No cross-agent shared graph. MCP protocol is not a first-class integration target. Each agent maintains its own isolated memory silo.; Single-user, single-agent model. No MCP-native interface. No shared workspace memory for agent teams..

Features2 agent-ready prompts

MCP-native graph write/read API that persists agent decisions and context across sessions and agent boundaries

▶

Team workspace sync layer with deduplication, conflict resolution, and per-member access control across multiple agents

▶

Competitive LandscapeFREE

Product	Does	Missing
mem0	Open-source memory layer for single AI agents. Raised $24M Series A. Pricing $19-$249/mo.	No cross-agent shared graph. MCP protocol is not a first-class integration target. Each agent maintains its own isolated memory silo.
Zep	Long-term memory for LLM apps. Per-user temporal knowledge graphs. Targets single assistant flows.	Single-user, single-agent model. No MCP-native interface. No shared workspace memory for agent teams.
Letta (formerly MemGPT)	Stateful agent framework with external memory storage. Requires using Letta's agent runtime.	Framework lock-in, only works with Letta agents. Not an open MCP server any agent can connect to.
Supermemory	Personal memory layer for ChatGPT and AI assistants. Consumer focus.	Consumer personal assistant use case only. No dev API, no cross-agent graph, no MCP server.