Connect Crosmos to AI clients with the Model Context Protocol.
Crosmos MCP is an optional service for AI clients that support the Model Context Protocol. It gives agents callable tools to search memories, add memories, list spaces, and check API health.Use MCP when you want callable Crosmos memory tools in an AI client such as Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, opencode, Cline, Roo-Cline, Zed, or Kimi CLI. Use the automatic-context plugins when you want memory to run through a host’s lifecycle, or use the SDKs when you are building Crosmos into an application or backend service.
The setup command authenticates with your Crosmos API key, detects installed MCP clients, writes their configuration, and can install the Crosmos skill for supported AI editors.It detects Claude Desktop, Claude Code, opencode, Cursor, VS Code, Windsurf, Cline, Roo-Cline, Zed, and Kimi CLI when they are installed.
Generate an API key from Settings → API Keys in the Crosmos Console. Keys use the csk_ prefix.
npx @crosmos/crosmos-mcp auth login
The login command validates the key and saves credentials to ~/.crosmos/credentials.json with restricted file permissions.
CROSMOS_API_KEY overrides saved credentials. Use environment variables for CI, containers, or client configuration that should not read from ~/.crosmos.
If you do not use setup, point your MCP client at npx -y @crosmos/crosmos-mcp.
Claude Code
Claude Desktop
Cursor
opencode
VS Code
Windsurf
Kimi CLI
claude mcp add crosmos-memory -- npx -y @crosmos/crosmos-mcp
Add to ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, ~/.config/Claude/claude_desktop_config.json on Linux, or %APPDATA%/Claude/claude_desktop_config.json on Windows.
