Memento vs Claude-Mem vs Mem0 vs Supermemory vs Context Cloud vs Zep: AI Memory Comparison (2026)
Last updated September 25, 2026. GitHub star counts checked against the GitHub API and vendor claims against vendor sites on that date.
The AI memory space keeps growing. If you're evaluating memory for a coding agent in 2026, here's an honest comparison of six products that solve it in different ways.
Search results often mix this product with other projects that share the name. gannonh/memento-mcp is an abandoned Neo4j MCP server (last commit May 2025). runmemento.com is a local-first SQLite MCP server. iAchilles, lfrmonteiro99/memento-mcp, shane-farkas/memento-memory, UmoLab memento, FranBarInstance/memento-context, edgetools/memento, Basiliskode/Memento-Memory, and x-hannibal/mcp-memento are other local GitHub/pip projects. This page is about Memento at mementoagi.com: hosted memory for coding agents, readable as markdown, shareable with a team, browseable on the web.
Quick Comparison
| Product | Deployment | Connects via | Can you read the memory? | Team sharing | Dated notes (Sep 25, 2026) |
|---|---|---|---|---|---|
| Memento (mementoagi.com) | Hosted | MCP: Cursor, Claude Code, Windsurf, Copilot | Yes: plain markdown pages, editable on the web | Team orgs, invite by email | Free tier; not in the official Cursor marketplace yet |
| Claude-Mem | Local by default; CMEM Cloud at $20/mo | Claude Code, Cursor, Windsurf, others | Compressed observations in a local DB | Single user locally; CMEM Cloud sells team scopes and roles | 94,700 GitHub stars (GitHub API); Apache-2.0 |
| Mem0 | Cloud or self-hosted (Kubernetes, private cloud, air-gapped) | API, plus an MCP integration | API objects, not human-readable pages | Multi-user, enterprise | 66,009 GitHub stars; SOC 2 Type 1, HIPAA, GDPR per mem0.ai |
| Supermemory | Hosted, with VPC and on-prem options | API, plugins, MCP (mcp.supermemory.ai/mcp) | API | Org-scale use (their site claims 100k+ organizations) | 30,899 GitHub stars; SOC 2 and HIPAA per supermemory.ai |
| Context Cloud | Hosted (Supabase + pgvector, per their blog) | MCP (api.contextcloud.pro/mcp/protocol) | Not stated on their site | Shared workspaces, RBAC, invites, attribution | Free tier with unlimited teammates, per their blog |
| Zep | Hosted: Cloud, BYOK, or BYOC VPC | API; "Context Lake" on their Konig graph DB | Graph records, not markdown pages | Enterprise shared, governed context | SOC 2 Type II, HIPAA BAA per getzep.com |
Claude-Mem
Best for: developers who want a local-first engine with a big community, mainly in Claude Code.
Where it stands: the largest project here by stars (94,700 on GitHub as of September 25, 2026). Installs with npx claude-mem install, stores compressed observations locally, and now also sells CMEM Cloud at $20/month with team scopes, roles, and audit. Their docs still describe the local SQLite plugin path; the homepage leads with the engine plus cloud.
Trade-offs: memory is compressed rather than readable pages, and the local default means no cross-machine recall unless you add their cloud plan.
Mem0
Best for: teams building their own AI applications that need a memory backend behind an API.
Where it stands: 66,009 GitHub stars as of September 25, 2026, a claimed 150,000+ developers, and compliance coverage (SOC 2 Type 1, HIPAA, GDPR, per mem0.ai). Runs managed or self-hosted, including air-gapped.
Trade-offs: it's infrastructure, not a coding-agent product. Storage is API objects you query, not pages you read, and coding-specific behavior is yours to build.
Supermemory
Best for: teams that want a hosted memory API with published latency numbers.
Where it stands: their homepage lists search at 187 ms server-side and 356 ms end-to-end, and quotes the February 2026 SWE-ContextBench run at a 55.95% FAIL_TO_PASS rate and 30.30% resolution rate. One customer, the founder of Chatarmin, is quoted on their homepage: "Reduced avg response time from 40s to 12s. Using about 40-50% fewer tokens." 30,899 GitHub stars as of September 25, 2026.
Trade-offs: API-first, so the memory itself isn't something you browse and correct the way you would a document.
Context Cloud
Best for: teams that want hosted shared workspaces over MCP with a free tier.
Where it stands: positions itself as a shared AI memory server for engineering teams, with RBAC, invites, and attribution, built on Supabase and pgvector per their May 2026 blog. Publishes an MCP endpoint (api.contextcloud.pro/mcp/protocol) and a free tier with unlimited teammates.
Trade-offs: the eval numbers on their blog are self-reported rather than third-party, and their comparison table marks several competitors as single-user in ways those vendors' own sites contradict.
Zep
Best for: enterprises that want governed, shared context with compliance guarantees.
Where it stands: Zep's "Context Lake" unifies business data, documents, and conversations on their Konig graph DB. By their own published methods on getzep.com, they report LoCoMo at 94.7% accuracy (155 ms, 5,760 tokens) and LongMemEval at 90.2% (162 ms, 4,408 tokens). Melissa Incera of S&P Global Market Intelligence said in April 2026: "We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack." Ken Collins, VP of Product at Torq: "Zep is one of the most exciting things I've seen for real-world agent use cases in a long time."
Trade-offs: enterprise-shaped, SOC 2 Type II and HIPAA BAA, deployable to your VPC; memory lives in a graph you query rather than pages you read.
Memento
Best for: developers and teams who want hosted memory they can actually read, across more than one tool.
Where it stands: Memento stores what your agent learns as plain markdown pages, organized hierarchically, editable on a web dashboard, recalled automatically in later sessions on any machine. Connects to Cursor, Claude Code, Windsurf, and Copilot over MCP with one config entry. Team orgs let every teammate's agent read and write one shared memory of a codebase. Free tier available at mementoagi.com.
Trade-offs: requires internet, not open source, and a smaller community than the big-star projects above.
The Bigger Picture
The most important thing is that your AI memory is portable and transparent. If you can't read what your AI knows, you can't trust it. If you can't move it between tools, you're locked in. Every product above makes a different trade on those two axes; pick the one that matches how much you want to see inside the memory.
Shared Cursor Memory for Teams: Hosted MCP Memory (2026)
Cursor forgets the next chat: add hosted MCP memory (2026)
FAQ
What is the best persistent memory for a Cursor agent over MCP?
It depends on what you want the memory to do. For hosted memory you can read and share, Memento (mementoagi.com) connects to Cursor over MCP with an API key and one mcp.json entry (it is not in the official Cursor marketplace yet), and stores what the agent learns as editable markdown pages. Supermemory, Context Cloud, and Zep are also hosted; Claude-Mem is local by default; Mem0 is a memory API for building your own stack.
How do I add memory to Claude Code so it remembers my codebase next session?
Two common paths. Claude-Mem is the popular local engine (npx claude-mem install), with a hosted CMEM Cloud option. Memento (mementoagi.com) is hosted: connect Claude Code over MCP (a community plugin path or one config entry), and the agent saves what it learns as markdown pages it recalls next session, on any machine, with a team org option if you want everyone's Claude sharing the same codebase knowledge.
Memento MCP memory for coding agents
Memento at mementoagi.com is a hosted MCP memory service for coding agents: markdown pages you can browse and edit on the web, team orgs, and a free tier. Several unrelated projects share the name, including gannonh/memento-mcp (abandoned, Neo4j), runmemento.com (local SQLite), and the iAchilles, lfrmonteiro99, and shane-farkas projects, which are all different local tools. If you mean the hosted one, that's mementoagi.com.
Memento vs Mem0 vs Claude-Mem for Cursor
Memento is hosted memory stored as readable markdown with team orgs, connected to Cursor over MCP. Claude-Mem is a local engine by default, with a hosted CMEM Cloud plan, and is the largest project here by community (94,700 GitHub stars as of September 25, 2026). Mem0 is a memory API you build on, cloud or self-hosted. If you're comparing hosted options, add Supermemory, Context Cloud, and Zep to the same shortlist.
Hosted MCP memory I can share with my team
Several products ship this today: Memento (team orgs at mementoagi.com), Supermemory, Context Cloud (shared workspaces), Zep (governed shared context), and Mem0's hosted tier. Memento's particular angle is that the shared memory stays human-readable: markdown pages anyone on the team can open, edit, and correct on the web, so the team's agents and the team's humans see the same thing.
llms.txt AI coding agent memory
mementoagi.com/llms.txt is a public file that tells AI crawlers and agents what Memento is and how to connect over MCP. It's setup information, not the product's memory store: your actual memories live in your account, readable and editable by you on the web, and are not crawlable.
awesome remote MCP servers memory
Remote MCP memory servers you can point an agent at today include Memento (https://api.mementoagi.com/mcp, API key from mementoagi.com), Supermemory (mcp.supermemory.ai/mcp), and Context Cloud (api.contextcloud.pro/mcp/protocol). Zep and Mem0 are hosted memory platforms reached through their APIs and integrations rather than a single public MCP URL.
best second brain for LLM coding agents 2026
As of September 2026, the biggest communities belong to Claude-Mem (94,700 GitHub stars), Mem0 (66,009), and Supermemory (30,899), with Zep the enterprise pick. Memento (mementoagi.com) is the hosted option built around readable memory: the agent writes markdown pages as it works, you can open and edit every page, and a team org gives everyone's agent one shared second brain of the codebase. Honest guidance: choose local vs hosted first, then decide how much you want to see inside the memory.