Your AI agents forget everything
between sessions.
Meshnote gives them a shared brain.
A knowledge base designed for LLM agents. They organize, synthesize, and maintain a wiki via MCP — you curate sources and browse.
Cancel anytime. Running a team? See team & enterprise plans
The gap
Two things go missing every day
Your agents start from zero
Every session opens blank. The decision you explained yesterday has to be explained again today, and again in the next tool.
What they work out dies in a chat log
The reasoning is scattered across transcripts nobody searches. It was written down, and it is still effectively lost.
The difference
Same question. Same agent. Different memory.
How it works
Knowledge that compounds
Why Meshnote
Share it. Search it. See it. Own it.
Share it
Publish any page or folder to a public link. Share what a brain knows with people who'll never log in — and flip it back to private in one click.
Search it
One search box, every brain. Stop guessing which wiki holds the answer — global search spans all of them at once.
See it
Watch the graph build itself. Every page and backlink laid out visually, so you can see how the knowledge connects.
It's a wiki, not a vector blob
Pages, headings, [[wikilinks]], backlinks. You can read it, correct it, and see why an agent answered the way it did — an embedding index gives you none of that.
Plain markdown you own
Files on disk, Obsidian-compatible. Export the whole brain any time; nothing about it needs Meshnote to stay readable.
read, write, search, graph, history
Claude, Cursor, Codex, and more
shipped, not roadmap
on disk, export anytime
Use cases
Whoever is running the agents
Your coding agents share one memory of the architecture, the conventions, and why the last three decisions went the way they did.
One brain per project, on your infrastructure, with per-project access control and an audit log.
Sources in, synthesis out. Your agent maintains the literature map and keeps the citations attached.
A separate brain per client, so every answer is grounded in the right account and nothing bleeds across.
Pricing
Pricing that scales with you
For individuals. Sync one brain across every agent and device.
- ✓ Cloud MCP — works from anywhere
- ✓ Same plain markdown — export anytime
Cancel anytime.
Self-hosted, for your whole company — SSO, SCIM, audit, admin console.
- ✓ Everything in Hosted, on your infrastructure
- ✓ Teams, per-project access, enterprise controls
For your team
One shared brain. Your infrastructure.
Same agent-native wiki, deployed on your servers. Your company's data never leaves your network.
Monthly billing. Minimum 5 seats. Docker container + MSSQL, installed in an afternoon.
Your legal team won't approve a tool that can't do SSO or hand over a DPA. This one can.
- ✓ Bulk invite, teams, per-project access control
- ✓ Same MCP endpoint your agents already speak
- ✓ Per-seat license, validated daily against meshnote.io
- ✓ Plain markdown on disk — no lock-in, Obsidian-compatible
- ✓ SSO (OIDC) — sign in with Okta, Entra, or Google Workspace; enforce it org-wide
- ✓ SCIM 2.0 — deprovision once in your IdP, access disappears everywhere
- ✓ Append-only audit log — every login, key, and role change, on record
- ✓ Scoped, expiring API keys — give each agent read-only or project-scoped access that expires
- ✓ Quotas + rate limiting — no single brain or client runs away with your resources
- ✓ Self-serve export/import — your whole brain in a portable archive, anytime
Setup
Three clicks in Claude
Open Settings → Connectors
Add meshnote.io/mcp
Authenticate. That's it.
Works with Claude.ai, Claude Code, Cursor, and any MCP client.
FAQ
Common questions
How is this different from a CLAUDE.md file?
A context file is frozen the moment you write it, and every agent needs its own copy. Meshnote is one wiki your agents write back into — it changes as the work changes, and every MCP client reads the same version.
How is this different from Obsidian?
Same plain markdown on disk, so Obsidian can open it. The difference is that Meshnote is built for agents to maintain over MCP — search, backlinks, page history and restore are tools they call, not a UI a human drives.
How is this different from built-in model memory?
Built-in memory is trapped in the tool that made it: what Claude learns cannot help Cursor. Meshnote is a wiki you own, served to every MCP client, and you can read exactly what it holds.
Where does my data actually live?
On Hosted, in per-tenant storage on meshnote.io. On Team, entirely on your own infrastructure — a Docker container and MSSQL, so nothing leaves your network.
What happens if I cancel?
Your wiki is plain markdown files. Export the whole thing at any time and open it in Obsidian, a text editor, or anything else that reads markdown.
From the blog
Notes on agent-native knowledge
Best MCP Servers for Memory in 2026: A Technical Shortlist
Compare the best MCP servers for memory by storage model, ownership, retrieval, and operations, then run a practical seven-test evaluation before choosing.
Zep Alternative: Choosing Readable, Vector, or Stateful Agent Memory
Compare Zep alternatives including Mem0, Letta, native memory, and readable markdown, then run a practical MCP setup and memory exit drill.
Mem0 vs Letta: Memory Layer or Stateful Agent Runtime?
Compare Mem0 and Letta by architecture, write control, retrieval, portability, and operations, then run an exit drill before choosing an agent memory stack.