Quickstart

You will connect a workspace, inspect available tools, and run a real Harbor Cloud execution. The CLI path here uses only commands that exist: hrbr help, hrbr inspect, hrbr exec, hrbr update, and hrbr serve mcp.

Go to tryharbor.ai, continue with Google or GitHub, then create your first workspace.

Install the CLI

npm i -g @zonko-ai/harbor
hrbr help

Start or verify local auth through inspect:

hrbr inspect 'return await hrbr.auth.status()'
hrbr inspect 'return await hrbr.auth.start()'
hrbr inspect 'return await hrbr.workspace.current()'

Connect a plugin

Use the dashboard Install page for the guided flow. After connecting a service, verify readiness from the CLI:

hrbr inspect 'return await hrbr.sources.list()'
hrbr inspect 'return await hrbr.tools.search({ query: "linear issues", limit: 3 })'

If a source requires OAuth, inspect can start that setup flow:

hrbr inspect 'return await hrbr.sources.oauthStart({ namespace: "linear-mcp" })'

Run a tool

Use the namespace and call shape returned by hrbr.tools.search:

hrbr exec 'return await linearMcp.listIssues({ limit: 3 })'

The call runs in Harbor Cloud, returns TOON-formatted output, and records a trace.

Point your agent at Harbor

For local MCP clients that launch a subprocess, use:

hrbr serve mcp

Hosted clients can connect to https://mcp.tryharbor.ai/mcp when they support remote MCP.

Where to go next

Browse plugins

Find the connector for the service you use.

Your first Function

Promote a working exec snippet into a reusable Function.

CLI reference

Learn the inspect/exec split.