---
title: "AgentDrive vs Memory MCP"
description: "The Memory MCP server keeps a local knowledge graph of facts an agent saves. AgentDrive syncs whole coding-agent sessions so another agent can resume them."
canonical: "https://agentdrive.sh/compare/memory-mcp"
---
# AgentDrive vs Memory MCP

> For the complete documentation index, see [llms.txt](https://agentdrive.sh/llms.txt). Every docs page is also served as markdown at its URL plus `.md`, and the full docs are in one file at [llms-full.txt](https://agentdrive.sh/llms-full.txt).

> The Memory MCP server keeps a local knowledge graph of facts an agent chooses to save. AgentDrive syncs whole coding-agent sessions so another agent can resume the work.

Source: https://agentdrive.sh/compare/memory-mcp. Last checked 2026-11-18. All comparisons: https://agentdrive.sh/compare.md.

## When to pick which

Pick Memory MCP when: You want an agent to keep a small set of facts, such as preferences or project entities, in a local file across sessions.

Pick AgentDrive when: You want the next coding agent, on this machine or another, to resume a task from what the last session did.

They store different things: an agent can keep durable facts in the Memory server's knowledge graph and pick up task state from AgentDrive.

| Feature | AgentDrive | Memory MCP |
| --- | --- | --- |
| Unit of storage | Sessions and files, versioned, with provenance | Entities, relations and observations in a knowledge graph |
| Where it runs | Hosted workspace plus a local CLI and MCP server | Local process over stdio (npx or Docker), or a Streamable HTTP service |
| Coding-agent sessions | Synced by hooks in Claude Code, Codex, OpenCode, Cursor and Antigravity; resumable from any of them | Not synced; the agent saves facts with tool calls |
| Storage location | AgentDrive R2 by default; your own R2 or S3 bucket on Team | A memory.jsonl file, set with MEMORY_FILE_PATH |
| Availability | Free plan; Pro $20/mo, Team $30/seat/mo | Open-source reference server (MIT, per its README) |

## About Memory MCP

- Described as "a basic implementation of persistent memory using a local knowledge graph" and published as `@modelcontextprotocol/server-memory`.
- Entities have a name, a type and a list of observations; relations are directed links between entities; observations are atomic facts stored as strings.
- Tools include `create_entities`, `create_relations`, `add_observations`, `read_graph`, `search_nodes`, `open_nodes` and matching delete tools.
- The graph is stored in `memory.jsonl` in the server directory unless you set `MEMORY_FILE_PATH`. The docs recommend an absolute path, because under npx the default sits inside `node_modules`.
- It talks over stdio by default, runs with npx or the `mcp/memory` Docker image, and can run as a Streamable HTTP service. Its README says it is licensed under MIT.
- The servers repository is moving from MIT to Apache-2.0: new contributions are Apache-2.0, and existing code stays MIT until its authors agree to relicense.
- The reference servers "are meant to serve as educational examples for developers building their own MCP servers", not production-ready solutions.

## Where Memory MCP works well

- Letting an agent remember a few facts, such as your preferences or the parts of a project, across sessions with no account.
- A simple, inspectable data model in one local file.
- Working with any MCP client.

## What AgentDrive does differently

- AgentDrive syncs the whole session from the agent's transcript. Nothing depends on the agent deciding to save a fact.
- The next agent gets a handoff brief with the goal, the files touched and the recent turns.
- The workspace is hosted, so an agent on another machine reads the same sessions and files.
- Files are versioned and content-addressed, every write keeps the name of the agent that made it, and `drive_search` searches by meaning.

## Frequently asked questions

### What is the difference between Memory MCP and AgentDrive?

The Memory MCP server keeps a local knowledge graph of facts an agent chooses to save. AgentDrive syncs whole coding-agent sessions so another agent can resume the work.

### When should I use Memory MCP?

You want an agent to keep a small set of facts, such as preferences or project entities, in a local file across sessions.

### When should I use AgentDrive instead?

You want the next coding agent, on this machine or another, to resume a task from what the last session did.

### Can I use AgentDrive alongside Memory MCP?

Yes. They store different things: an agent can keep durable facts in the Memory server's knowledge graph and pick up task state from AgentDrive.

## Sources

- https://github.com/modelcontextprotocol/servers/tree/main/src/memory
- https://github.com/modelcontextprotocol/servers/blob/main/LICENSE
- https://github.com/modelcontextprotocol/servers

