APIs
MCP is one way in. When you're building your own agents, services, automations, and plugins, Grafana AI also exposes APIs you can call directly, so the Assistant becomes a component in your systems, not just a destination.
Each path gives your code a structured way to ask Grafana for context and hand work back, using your Grafana identity and permissions.
Agent-to-agent (A2A)
Let your own agents talk to Grafana Assistant over the open A2A protocol. Delegate a focused task (analyse a signal, draft a query, summarise an incident) and get structured results back that your agent can act on. Grafana AI participates as one agent in a wider multi-agent workflow instead of owning the whole loop.
Webhooks
Wire Grafana AI into the events that already drive your work. Trigger an agent run from an alert, a deploy, or a pipeline step, and receive structured results back at an endpoint you control, ready to post to chat, open an issue, or kick off the next step.
HTTP APIs
Call the Assistant programmatically from scripts, services, and CI. Send a prompt with the context you already have, run a saved skill, and read back the result: the same grounded work the Assistant does in Grafana, available where your automation runs.
Frontend API
Building Grafana plugins or apps? The @grafana/assistant TypeScript SDK puts
the Assistant inside your own UI. Open it with a prompt and the context
your page already knows, drop in ready-made buttons and prompt cards, and
generate content inline in your forms.
Prefer to work inside an editor or MCP-compatible client instead? See MCP integrations.