Documentationbreadcrumb arrow Grafana Cloudbreadcrumb arrow AI and machine learningbreadcrumb arrow Grafana Assistantbreadcrumb arrow Playbooksbreadcrumb arrow Trigger actions with MCP calls
Grafana Cloud

Trigger actions with MCP calls

Note

Preview feature: Playbooks are currently in preview. Functionality may change in future releases.

When you run a playbook in Assistant, you can instruct the agent to make MCP (Model Context Protocol) calls that trigger actions in external systems. For example, an agent can create a GitHub issue, send a notification, update a ticket, or trigger a workflow—all based on what it discovers during an investigation.

Note

Assistant only: MCP calls in playbooks only work when you run playbooks in Assistant. Currently, investigations don’t have access to MCP tools. If you run a playbook in another context, such as investigations, the agent cannot make MCP calls.

Note

Approval required: When an agent wants to make an MCP call, it asks for your approval first. You review the proposed action and approve or reject it. MCP calls don’t execute automatically.

How it works

When you write a playbook, you can include instructions that tell the agent to use MCP tools. When the agent runs the playbook in Assistant, it follows your instructions and proposes MCP calls when appropriate.

Here’s the workflow:

  1. You run a playbook in Assistant
  2. The agent executes the playbook steps
  3. When the agent needs to make an MCP call, it proposes the action to you
  4. You review the proposed MCP call and approve or reject it
  5. If approved, the agent executes the MCP call and continues with the playbook

Example: A playbook might tell an agent to:

  • Create a GitHub issue when an investigation finds a bug
  • Use Slack MCP to send notification when a new GitHub issue is created
  • Update a Linear ticket with investigation findings
  • Trigger a deployment workflow in another system

Before you begin

To use MCP calls in playbooks, you need:

  1. Assistant access: Run playbooks in Assistant. MCP calls don’t work in investigations.
  2. MCP integration: Set up an MCP server integration in Assistant Settings > Integrations
  3. Tools enabled: Make sure the specific MCP tools you want to use are enabled for the integration
  4. Permissions: Verify you have permission to use MCP integrations

Write playbooks with MCP calls

When you write a playbook that uses MCP calls, be clear about when and how the agent should use them.

Specify when to make calls

Tell the agent exactly when it should propose an MCP call:

Markdown 
After completing the investigation:

1. If a bug is found, use the GitHub MCP integration to create an issue
2. Include the alert name, root cause, and remediation steps in the issue
3. Assign the issue to the on-call team

Define conditions

Specify the conditions that trigger an action:

Markdown 
If the investigation identifies a bug:

- Use the Linear MCP tool to update ticket [TICKET-ID]
- Set the status to "In Progress"
- Add the investigation findings as a comment

Provide context

Include all the information the agent needs to make effective MCP calls:

Markdown 
When the alert severity is critical:

- Use the Slack MCP integration to send a message to #on-call
- Include the alert name, current status, and investigation summary
- Tag the on-call engineer

Best practices

Test MCP calls first: Before you add MCP calls to a playbook, test them manually in Assistant to make sure they work as expected.

Use specific conditions: Only trigger actions when specific conditions are met. This prevents unnecessary notifications or workflow triggers.

Include all details: Provide all the information the agent needs, such as ticket IDs, channel names, or user names.

Review proposals carefully: When an agent proposes an MCP call, review it before approving. Check that it matches what you intended.

Start simple: Begin with simple actions like creating an issue or sending a notification. Once those work well, you can build more complex workflows.

Remember approval is required: MCP calls don’t execute automatically. The agent asks for your approval each time, so plan your workflows accordingly.

Limitations

  • Assistant only: MCP calls only work when you run playbooks in Assistant. Currently, investigations don’t have access to MCP tools, so playbooks run in investigations cannot make MCP calls
  • Approval required: You must approve every MCP call before it executes. Agents cannot make MCP calls automatically
  • Integration must be configured: The MCP integration you reference must be set up and available
  • Tools must be enabled: Agents can only use MCP tools that are enabled for the integration
  • Handle failures: Playbooks should account for MCP call failures. Consider adding fallback instructions in case an MCP call fails

Next steps