Skip to main content

Overview

Run a Honeycomb MCP prompt to start a guided, multi-step workflow in your AI agent without writing the instructions yourself. A prompt is a set of pre-written instructions that your agent follows, often by calling several MCP tools in sequence. Your agent chooses tools on its own based on what you ask, while you run a prompt when you want a specific workflow.

Syntax

AI agents typically display MCP prompts as slash commands, such as /honeycomb-onboarding. The exact name can vary by agent. For example, Cursor displays /honeycomb-onboarding as /honeycomb/honeycomb-onboarding.
Honeycomb MCP prompts differ from commands in the Honeycomb plugin, such as /honeycomb-setup. Plugin commands exist only when you install the plugin. MCP prompts come from the Honeycomb MCP server, so they work with or without the plugin.

Before you begin

Prompts run through your Honeycomb MCP connection, so set up that connection first. Make sure that:
  • Your AI agent connects to Honeycomb MCP and authenticates. To set up a connection, visit Connect to Honeycomb MCP.
  • Your AI agent supports MCP prompts.
Prompts don’t require any additional scopes.

Onboarding and learning prompts

These prompts help new users get data into Honeycomb and learn how to work with it.
Guides you through getting data into Honeycomb. Your agent asks which path you want to follow:
  • Get real data into Honeycomb: Send telemetry from your app.
  • Send sample data: Generate sample data based on your codebase.
  • Analyze existing data: Explore data already in your environment.
On the real data path, your agent reviews your repository and chooses a next step based on what it finds:
  • Already sending to Honeycomb: Your agent confirms that your events arrive.
  • Sending OpenTelemetry data somewhere else: Your agent points your existing exporter at Honeycomb.
  • Sending data to Datadog or New Relic: Your agent adds an OpenTelemetry Collector alongside your existing agent, so Honeycomb receives a copy of your data while your current vendor keeps receiving it. If you prefer, your agent can replace your vendor instrumentation with OpenTelemetry instead.
  • Little or no instrumentation: Your agent offers to add OpenTelemetry to your app. With the Honeycomb plugin installed, your agent instruments your app for you. Without the plugin, your agent guides you through instrumenting your app manually.
Once your first events arrive, your agent helps you start a Canvas investigation.
Gives you a guided tour of your Honeycomb workspace. This prompt doesn’t instrument your app or send data. Your agent asks which path you want to follow:
  • Exploring: Environments, datasets, columns, queries, SLOs, and Triggers.
  • Debugging: Heatmaps, traces, and BubbleUp.
  • Reliability: SLOs and error budgets.

OpenTelemetry instrumentation prompts

These prompts help your agent evaluate and improve the OpenTelemetry instrumentation in your code. Each prompt accepts either a code snippet or a path to a code file. When you provide a file path, your agent reads the file itself, so it needs access to that file.
Asks your agent to explain your code and suggest OpenTelemetry automatic instrumentation, with links to the OpenTelemetry documentation for your language. This prompt produces guidance and doesn’t change your code.Arguments: Provide either code_snippet (the code as text) or file_path (the path to a code file).
Asks your agent to add custom OpenTelemetry spans to your code.Arguments: Provide either code_snippet (the code as text) or file_path (the path to a code file).

Troubleshooting

Running into issues? Here are some common problems and ways to fix them.
Still stuck? Visit the Support Knowledge Base or post a question in the Pollinators Community.

Prompts don’t appear in your agent

When Honeycomb MCP tools work but a prompt like /honeycomb-onboarding doesn’t appear, the cause is usually how your agent names or loads prompts. Check the following:
  • Look for a different name: Some agents add a prefix to prompt names. For example, Cursor displays /honeycomb-onboarding as /honeycomb/honeycomb-onboarding.
  • Confirm your agent supports MCP prompts: Agents that support only MCP tools don’t display prompts.
  • Check your toolset header: If you configured the X-MCP-Toolset header, set it to all or query, or remove it. A value of canvas doesn’t include prompts. To learn more, visit Connect to Honeycomb MCP: Limit the tools your agent loads.

Next steps