Cozy vs Logseq: choosing a Logseq alternative for agents

Updated · By the Cozy team

Cozy is a hosted Markdown wiki with one hosted MCP endpoint per account, scoped agent access and revision history. Logseq is an open-source local outliner built around blocks. Choose Logseq if you want a local graph you fully control. Choose Cozy if you want your own agents to read and write one wiki without a desktop app running.

We make Cozy, so read this with that in mind. Logseq is a mature, loved tool, and for many people it is the better choice. This page covers what differs, where Logseq is ahead, and when a hosted wiki makes more sense.

The short version

Logseq is a local-first outliner. You write in blocks, link pages and blocks, and keep your graph on your own machine. It is open source under AGPL-3.0 per its GitHub repository.

Cozy is a private wiki and cloud file space for each person and their AI agents. Content is Markdown, text, JSON and CSV files. Cozy runs no model itself: you bring your own agent and connect it over MCP.

QuestionLogseqCozy
Where data livesOn your machineHosted on Cloudflare
Unit of contentBlocks in an outlineWhole Markdown files
Agent accessLocal MCP server in the desktop appOne hosted MCP endpoint per account, optional per-workspace endpoints
Agent permissionsBearer tokenOAuth scopes, or credentials with folder prefixes and expiry
Source codeOpen source, AGPL-3.0Hosted service

Agent access: local MCP versus hosted MCP

Logseq has a built-in MCP server. Per Logseq issue 12413, you turn on "Enable MCP" under Settings, Features, and the desktop app serves Streamable HTTP at http://127.0.0.1:12315/mcp with bearer-token auth. The app has to be open on the same machine as the client. That issue, a 400 "No valid session ID" error, was opened in February 2026 and was still open when we checked, though a contributor says a fix is on master.

Community servers such as ergut/mcp-logseq also need Logseq running with its HTTP API server and a token. They offer a broader tool set, and everything stays local.

Cozy exposes one hosted MCP endpoint per account at https://cozy.cloud/mcp. It is stateless Streamable HTTP, so any MCP client can reach it with no desktop app running, and it covers all your workspaces. See the MCP server guide for setup. In Claude Code or Codex:

claude mcp add --transport http --scope user cozy https://cozy.cloud/mcp
codex mcp add cozy --url https://cozy.cloud/mcp

Both sign in through Cozy's consent page with OAuth. For scripts and headless agents, each workspace also has an endpoint at https://cozy.cloud/wiki-mcp/WORKSPACE_UUID that takes a bearer credential.

If your agent runs on your laptop next to Logseq, the local route works and costs nothing. If your agents run in other places, or you want one wiki that several agents share, a hosted endpoint is simpler.

Safety when agents write

Letting an agent edit an outliner is risky. The matryca-plumber project exists because agents editing Logseq Markdown directly can cause damage. It adds a shadow search index, optimistic concurrency control, page locks and tiered permissions as a separate daemon on one machine.

Cozy builds comparable controls into the service:

Structure: blocks versus files

Logseq's block model is its signature. You can reference and embed a single block anywhere. Cozy does not offer block-level addressing. A page is a file, and agents read and write whole files. That avoids the problem reported in Logseq issue 12220, where early MCP output exposed only top-level blocks. It also gives up block references and embeds, which many Logseq users rely on.

Cozy derives its graph from Markdown conventions on every save. Frontmatter fields such as type, description, aliases and tags are read. Frontmatter keys whose values are wikilinks become typed relations. You can also write relation:: [[target]] lines. Links into sources/ count as citations. Read more in typed links and Markdown wiki.

Where Logseq is better

Be aware of Logseq's transition too. The repository README says the DB version is beta, warns that data loss is possible, and recommends backups and a test graph. Sync for the beta is tied to a backer or sponsor level on OpenCollective. The file-based line and the DB version are different formats, so check which one you use before you plan a migration.

Where Cozy fits

Cozy also supports a "dreaming" pass that your own agent runs: it checks recent changes against sources, records CONFLICT: lines and merges duplicates. Cozy provides the tools and prompts; the quality depends on your agent.

Moving from Logseq

Cozy accepts Markdown, so Logseq pages can be brought over as files. We have not verified that a Logseq import is lossless, so treat block references, embeds and properties as items to check by hand. The Logseq import guide covers the current path. You can also keep Logseq for daily writing and push selected pages into Cozy through the ingest API for agents to use.

How to choose

Pick Logseq if local ownership, open source and block-level outlining matter most, and your agent runs on the same machine. Pick Cozy if you want hosted, governed agent access to a plain-file wiki. Many people can reasonably use both. For other options, see the comparison hub.

Frequently asked questions

Is Cozy a Logseq alternative?

For agent-heavy work, yes. Cozy is a hosted Markdown wiki with a hosted MCP endpoint, while Logseq is a local, open-source block outliner. If you rely on block references, journals and offline use, Logseq is still the better fit.

Does Logseq have an MCP server?

Yes. Logseq has a built-in MCP server you enable in the desktop app, served locally at 127.0.0.1:12315/mcp, so the app must be running. Cozy's MCP endpoint is hosted, one URL per account at https://cozy.cloud/mcp. See the MCP server guide.

Can I import my Logseq graph into Cozy?

Cozy stores Markdown, so pages can be added as files. We have not verified a lossless import, so check block references, embeds and properties by hand. See the Logseq import guide.

Is my data open source or local with Cozy?

No. Cozy is a hosted service on Cloudflare, operated by Coral Sourcing LLC. Your content is plain Markdown, text, JSON and CSV, but Cozy does not offer Logseq's local, open-source model. See privacy.

Give your agents a memory worth keeping

Cozy gives every account a private wiki that Claude, Codex, ChatGPT and other MCP clients can read and maintain.

Start your wiki