Connect your agent

You do not need an AI agent to use Kreel. The dashboards are a first-class way to work with your data on their own. Connecting an agent adds a second way to work with the same data: ask questions in plain language, run a repeatable report, audit an account, or draft ad copy, without leaving your chat client, terminal, or editor.

Every connected agent reads the same data as the dashboard, and shares the same Brand Knowledge Base that keeps a brand’s positioning, targets, competitors, and past learnings around between conversations. An agent’s writes are limited to Kreel’s own resources: the Brand Knowledge Base and hosted reports, plus explicit drafts you paste into your ad or email platform yourself. It never changes a live campaign, flow, product, or customer record.

Three ways to connect

Pick the route that matches the agent you use.

MCP is the recommended route for chat agents like Claude (web, desktop, and Cowork) and ChatGPT (Business and Enterprise plans, with Developer mode enabled). You add Kreel’s MCP server as a custom connector with one URL and approve a login popup, so there is no API key to paste and no config file to edit. See Connect over MCP.

The kreel CLI is for scripts, terminal-based agents, and automation such as CI. You authenticate with an API key once, then run commands like kreel summary or kreel query directly. See The kreel CLI.

The Claude Code plugin ships Kreel’s ready-made prompts as /kreel:* slash commands inside Claude Code. It only works alongside an MCP connection, since its commands call MCP tools. See The Claude Code plugin.

A few setups (stdio-only MCP clients, ChatGPT Custom GPT Actions, scripts, and CI) authenticate with an API key instead of the MCP login popup. Create and manage those keys under API keys.

What your agent can do once connected

Once connected, an agent can pull cross-platform performance, entity lists, and creative analysis across Shopify, Meta, Google, and Klaviyo (see Data tools); run an open-ended query against the underlying tables when the built-in tools do not cover what you need (see Query); read and update the brand’s persisted profile, positioning, competitors, and past learnings so context survives between conversations (see Brand context tools); and publish a frozen, password-protected report to the brand’s hosted-reports subdomain (see Publish a report). It can also run one of Kreel’s packaged prompts, like a weekly report or a creative fatigue check, instead of building a request from scratch (see Prompts and resources and the prompt library).

How brand selection works

Kreel groups your data by brand (one connected store or account) inside a workspace. See Workspaces, brands, and seats for how those fit together. A logged-in MCP connection reaches every brand across every workspace you belong to. An API key reaches every brand its creator can access, but only within the single workspace the key was created in.

Every brand-scoped MCP tool takes an optional brand argument, matched by name or id. CLI commands take a --brand flag with the brand’s id.

• If you can reach exactly one brand, it is used automatically and you can leave it out.
• If you can reach more than one, you must pass it.
• Ask your agent to list the brands you can reach (or run kreel brands) to see their names and ids.

This is checked fresh on every call rather than stored on the connection or key, so if your access to a brand changes, what the agent can reach changes with it.

Need help?

Email support@kreel.ai.