Atlaso Labs MCP

What the Atlaso Labs MCP MCP server does

Atlaso Labs MCP MCP server connects MCP-compatible AI clients to Atlaso's hosted memory service. It keeps durable notes such as decisions, preferences, and project context available across connected tools. Personal memories follow the user, while project memories are associated with the repository or workspace where they were created.

The service is intended for situations where an agent needs continuity beyond a single conversation or application. Atlaso recalls relevant context before a turn and can capture durable information afterward. Recall results include a verdict indicating whether a memory is settled or contested, helping the model distinguish established information from uncertain context.

How it works

The server is accessed remotely at https://mcp.atlaso.ai/mcp using Streamable HTTP. Authentication uses OAuth 2.1 with dynamic client registration. The MCP client handles browser authorization, and the connection does not require an API key pasted into configuration.

Atlaso Labs MCP MCP server exposes memory operations directly to the client. recall searches stored notes and returns ranked results, remember creates a durable note with a polarity, and recent lists the newest memories. forget permanently removes a memory by ID, while status reports connection state, memory counts, and a memory health score.

The repository documents the hosted service rather than providing server code to run locally. Terminal integrations can also be installed with Atlaso's shell or PowerShell installer, which sets up the atlaso CLI, links the machine, and lets the user select supported AI tools for shared memory.

Setup and configuration

Add the remote URL as an MCP connector in a compatible client and authorize in the browser when prompted. A generic configuration has this shape:

{
  "mcpServers": {
    "atlaso": {
      "url": "https://mcp.atlaso.ai/mcp"
    }
  }
}

Claude Desktop and claude.ai can use the custom connector settings. Cursor, VS Code, and other remote-MCP clients can use the URL in their MCP configuration. Claude Code, Codex, OpenCode, and Antigravity can use the provided installer for CLI-oriented hooks that add automatic recall and capture beyond the MCP connection itself.

No environment variables or API keys are documented as required. Account creation or machine linking occurs during authorization or the CLI setup process.

Tools and capabilities

• recall(query, limit=5): searches memory and returns ranked {id, content} results.
• remember(text, polarity): stores a note with positive, open, cautionary, or negative polarity.
• recent(limit=10): lists recent memories in newest-first order.
• forget(id): permanently deletes one memory and cannot undo the operation.
• status(): reports connection status, stored and pending memory counts, and a health score.

Secrets are scrubbed before storage. Data is encrypted in transit and at rest, and the README states that memory is not used to train AI models or sold. Individual memories or the entire account can be deleted.

Limitations and notes

Atlaso is a managed service; this repository does not contain self-hostable server code. Use therefore depends on access to the hosted endpoint and browser-based OAuth authorization. Memory scope is assigned automatically as personal or project context rather than being managed manually.

The free plan supports one device and one tool. Paid plans add broader device and tool sharing, with additional features varying by plan. The Build plan includes a developer memory API. forget is destructive, so clients should use it only when permanent deletion is intended.