Does Letta support MCP?

Yes, Letta supports MCP for connecting tools to its agents. The difference is altitude: Letta is an agent framework and runtime you build on, while Glen is a single MCP tool that provides org-shared memory with no runtime to adopt.

Is Glen open source like Letta?

Letta is open source under Apache 2.0 and self-hostable. Glen is a managed, MCP-native shared-memory service you connect to rather than operate, optimized for getting a whole org's agents on one memory store quickly.

Glen vs Letta

Glen and Letta (formerly MemGPT) both give AI long-term memory, but they operate at different levels of the stack. Letta is an open-source, self-hostable agent framework where the agent itself manages memory through tool calls, using an OS-inspired hierarchy of context and storage tiers — you build and run stateful agents on the Letta runtime. Glen is shared memory for AI agents delivered as a single MCP tool, scoped to an organization so that every agent on a team reads from and writes to the same store, with no agent runtime to adopt. This comparison contrasts them honestly on scope, integration, MCP support, and the open-versus-managed tradeoff, so you can match the model to how your agents actually work.

How they compare

Dimension	Glen	Letta
Memory scope	Org-shared by default: every agent in the organization reads and writes one store, so a fact one teammate's agent learns is available to all.	Per-agent memory the agent manages itself; sharing knowledge across a whole team is something you architect on top of the framework.
Integration model	MCP-native: a single glen tool any MCP client calls, with no agent runtime or framework to adopt.	An agent framework and runtime you build on; Letta also supports MCP for connecting tools, but the core model is running stateful agents on its platform.
Setup	Connect the MCP server once; new teammates inherit the org's existing memory, with no per-user store to seed.	Stand up the Letta server (self-hosted or managed) and build agents whose memory blocks they curate; powerful but more to operate.
Open vs managed	Managed, MCP-native shared memory you connect to rather than operate.	Open-source (Apache 2.0) and self-hostable, giving full control of data and infrastructure, with a managed cloud option too.

Verdict

Letta is a strong choice when you want to build and own stateful agents end to end: it is open source, self-hostable, and its self-managing memory hierarchy is genuinely sophisticated for a single agent's continual learning. But if your goal is for a whole team's agents to share what they learn — without adopting an agent runtime — Glen is built for exactly that. It delivers org-scoped memory as one MCP-native tool, with read and write in a single round trip and no per-agent silos to stitch together. Choose Letta to engineer self-managing agents you host; choose Glen when the unit you care about is the organization and you want shared memory that drops into the MCP clients your team already uses.

FAQ

Does Letta support MCP?: Yes, Letta supports MCP for connecting tools to its agents. The difference is altitude: Letta is an agent framework and runtime you build on, while Glen is a single MCP tool that provides org-shared memory with no runtime to adopt.
Is Glen open source like Letta?: Letta is open source under Apache 2.0 and self-hostable. Glen is a managed, MCP-native shared-memory service you connect to rather than operate, optimized for getting a whole org's agents on one memory store quickly.