CLAI

A desktop multi-agent orchestration app with MCP-native tools, local execution capabilities, shared workspaces, and scheduled automations.

Features

• MCP-Native Assistant - Attach MCP servers to a tab and the assistant uses their tools directly. Each tab scopes its own capability access.

• Scheduled Automations - Create automations that run on a schedule with their own MCP servers, filesystem grants, and shell policies.

• Local Execution - Agents get a private workspace directory and can optionally read/write additional paths and run shell commands, controlled by per-agent allow/block lists.

• Shell Access Modes - Three modes per agent: Off (no shell), Restricted (only explicitly allowed command prefixes can run), and Full (anything not blocked). Prefix matching supports subcommands — allowing kubectl get permits kubectl get pods but not kubectl delete.

• Run Notices - When an agent hits a policy denial, the run completes with an amber "warnings" state instead of failing silently. The Fleet view surfaces these so you can adjust permissions.

• Workspace Orchestration - The assistant can split tiles, add canvas nodes, open anomalies panels, and create dashboard charts as part of a workflow.

• Canvas and Dashboards - Automations and chat sessions create visual artifacts directly in the workspace with explicit targets.

• Tabbed Capability Scopes - Normal tabs start with no MCP access until you attach servers. Automation tabs inherit their MCP and execution config.

• Provider Flexibility - Connect any OpenAI-compatible provider and model, including OpenAI, together.ai, Groq, and local endpoints.

Important: local execution is not sandboxed. Shell commands run as your OS user with full privileges. The allow/block lists control which commands the agent can invoke, but do not restrict what those commands can access — if you allow cat, the agent can read any file your user can, not just the paths in the grant list. Path grants only apply to the built-in fs.read/fs.write tools, not to shell commands. LLMs can also chain commands in unexpected ways. Always use Restricted mode with a minimal allow list, and review run notices to catch anything you didn't anticipate.

Installation

Download the latest release for your platform from the Releases page:

Platform	Download
Windows	.msi or .exe
macOS	.dmg
Linux	.deb, .rpm, or .flatpak

Getting Started

1. Connect a provider - In Settings, add an OpenAI-compatible provider and choose a default model
2. Add MCP servers - Register local or remote MCP servers in Settings
3. Attach servers to a tab - Use the Add MCP badge in the tab context bar to grant tools
4. Chat or orchestrate - Ask the assistant to inspect tools, fetch data, open panels, or build visualizations
5. Create agents - Set up scheduled automations with their own MCP servers and local execution policies
6. Configure local execution - Grant filesystem paths and shell access per agent, starting with Restricted mode

Capability Model

CLAI scopes every capability explicitly:

• MCP servers are configured globally in Settings, then attached per tab or per automation
• Normal tabs have no MCP access until you attach servers
• Automations use the MCP servers and execution policy from their config
• Built-in tools handle workspace orchestration — tabs, canvas, dashboards, anomalies panels
• Local execution is per agent: a private workspace directory, optional extra path grants (read-only or read-write), and shell access with prefix-based allow/block lists
• Domain tools come from MCP servers

Provider Setup

CLAI uses a configured provider plus a default model for chat, Fleet interventions, and scheduled automations.

In Settings:

• connect an OpenAI-compatible provider using an API key
• optionally set a custom base URL for compatible providers
• choose the default model for assistant sessions and automations

Compatible with OpenAI, together.ai, Groq, local OpenAI-compatible endpoints, and other compatible providers.

Netdata Cloud Integration

Adding a Netdata Cloud MCP server unlocks extra workspace tools:

• Dashboard charts — the assistant can add live metric charts directly to the workspace
• Anomaly panels — open dedicated anomaly investigation views for any monitored node
• Canvas visualizations — build visual topologies with metric chart nodes, status badges, and markdown

These tools (dashboard.addChart, canvas.addChart, anomalies.open) appear automatically when a Netdata-capable MCP server is attached to the session.

Development

# Clone and install
git clone https://github.com/juacker/clai.git
cd clai
npm install

# Run the desktop app in development
make dev

# Verify builds
npm run build
cargo check --manifest-path src-tauri/Cargo.toml

Architecture

• Frontend: React + Tauri
• Workspace runtime: tabs, split tiles, canvas, dashboards, anomalies panels
• Assistant runtime: session-scoped built-ins + MCP tools, optional fs/shell access
• Automation runtime: scheduled agents with private workspaces and execution policies
• MCP support: per-server config, optional bearer auth, HTTP and stdio transports

License

MIT