Atlaso
Give coding agents durable, project-scoped memory on disk without shipping everything to a hosted memory SaaS.
Overview
Atlaso is a MCP server for the Build phase that stores agent memory locally in SQLite with conflict-aware recall and project-scoped storage.
What is this MCP server?
- Local SQLite store with conflict-flagging recall so stale facts surface instead of silently overwriting
- Scope-bag dispersion for organizing memories by project/context rather than one flat blob
- ATLASO_PATH or cwd walk-up resolution (pyproject.toml, package.json, .git) — never defaults to $HOME
- PyPI package atlaso via uvx with stdio transport (version 0.1.0a6)
- Open-source alternative to proprietary agent memory backends
- Server version 0.1.0a6 on PyPI identifier atlaso
- Stdio transport via uvx with optional --with fastmcp
- Store resolution uses project markers: pyproject.toml, package.json, .git, Cargo.toml, go.mod
Community signal: 8 GitHub stars.
What problem does it solve?
Coding agents forget repo-specific decisions every session, and cloud memory tools add cost and data-leak anxiety for solo builders.
Who is it for?
Privacy-conscious solo builders running Claude Code or Cursor on real repos who want local, versionable agent memory tied to project roots.
Skip if: Teams that need shared org-wide memory across many machines without sync, or builders who want zero local Python/uv setup.
What do I get? / Deliverables
After you register Atlaso in your agent, recall persists under your project’s .atlaso/ folder with flags when memories conflict so the agent can reconcile context instead of guessing.
- Project-local .atlaso/ SQLite memory store resolved from cwd or ATLASO_PATH
- MCP tools for recall with conflict-flagging behavior
- Agent sessions that retain scoped memories across restarts
Recommended MCP Servers
Journey fit
Agent memory is wired in during build when you configure MCP tooling and project context for Claude Code or similar clients. Canonical shelf is agent-tooling because Atlaso is an MCP stdio server that stores recall in local SQLite under a project-resolved .atlaso/ directory.
How it compares
Local MCP memory store with SQLite, not a hosted vector memory SaaS or a one-off prompt skill.
Common Questions / FAQ
Who is Atlaso for?
Solo and indie builders who want open-source, on-disk memory for coding agents without sending project context to a third-party memory API.
When should I use Atlaso?
Use it when you start a multi-session build or operate loop on one repo and need the agent to remember architecture choices, env quirks, and prior fixes with conflict awareness.
How do I add Atlaso to my agent?
Add a stdio MCP entry that runs uvx with the atlaso PyPI package (mcp subcommand), set ATLASO_PATH if you want a fixed store path, then reload MCP in Claude Code, Cursor, or Codex.