Loki is a local MCP server for AI-assisted development. It gives an MCP client controlled access to a development workspace for files, Git, isolated jobs, previews, artifacts, and optional integrations without giving project code the host's Docker socket or machine credentials.
Run in PowerShell:
irm https://jinyongp.dev/loki/install.ps1 | iexThe installer verifies the native release archive and installs the Loki CLI.
Then run loki setup. Choose individual tool groups; Loki downloads verified
releases, prepares their execution host and starts the selected services.
curl -fsSL https://jinyongp.dev/loki/install.sh | shThe installer installs only the native Loki CLI. Run loki setup afterward to
choose an execution host, manage tools and connect your client. Existing 0.1 installations use
their own lifecycle and must be removed before using the same command directory.
See installation and tool selection for supported hosts, manual selection, remote execution and remaining acceptance gates.
Browser-only tools can run on Linux/macOS and native Windows where supported.
Other tool groups use a Linux execution host. Windows setup offers a dedicated
loki-tools WSL distribution with creation confirmation; Linux setup can prepare
its managed Docker Engine service. An existing WSL Ubuntu or other distribution with Loki installed can be selected,
and an existing Linux host can also be prepared over SSH.
Add the Codex MCP connection with loki connections setup codex; browser-only
connections also take --workspace /absolute/project/path.
Run loki status for the current target, tool states, client configuration and
next actions. loki doctor verifies runtime readiness. loki tools manage opens
the tool menu directly; installation and activation are independent.
Guided setup and the restored operator commands are available from 0.2.6. See restoration tracking for native evidence and remaining intended-machine checks.