Create a Custom Service Health Board With the Honeycomb MCP

Your software is sending data to Honeycomb. Now where is the dashboard you want?

The best dashboard is one created just for your application, or your service, or your team. You can get that in minutes with the Honeycomb MCP.

Open your coding agent in your IDE, or on the command line in your code repository. Configure the Honeycomb MCP and authenticate with Read and Write permissions. Now tell it what you want. You can be high-level:

Make me a service health board for the frontend service.

Or you can get specific:

Make me a service health board for the frontend service. I want to see the latency and error rates by endpoint, and I want to see the top 10 users by latency. Also, I want to see a comparison of the latency and error rates for the last 24 hours, 7 days, and 30 days. And I want to see a list of all the dependencies, and their latencies and error rates. Also all SLOs related to this service. Put a text panel above each graph explaining it. Ask me clarifying questions. Print the plan for the board for my approval before you create it.

Prefer videos? Watch the demo below:

Custom boards are the best boards

Your agent combines knowledge of your code and your data to create a board that's exactly what you need.

From the code and the web, it can:

• Look at code to find service names and entry points
• Search the code for custom attributes, which are the high-value fields that don't show up on generic dashboards
• Check the programming language, and look up OpenTelemetry standard field names
• Guess at the business metrics that matter

Then, in Honeycomb, it can:

• See the schema of your data to find all available fields
• Look at existing boards and frequently-run queries to see what people find important
• List SLOs and triggers to see what's alerting on
• Notice the standard attributes AND the non-standard ones (which are often the interesting ones)
• Look for infrastructure metrics and choose them based on the attributes in your spans
• View traces for entry points and dependencies

Checking code, looking at real data, and asking you questions—it has all the context it needs to make something really useful. The Honeycomb MCP gives it tools to dig around in real data and then to create the board. For added thoroughness, download or paste in a skill like this one.

What you get

Your agent gives you a link to the custom board. Check it out. Click into the queries to see their full definition. For example, in the video posted above, the board included:

• SLOs at the top
• Key stats, like latency, errors, status codes
• Everything broken down by route
• Dependency services and their latencies
• Business metrics
• Top users

Once it's created, it's yours to adjust. You can click into any query to tweak it (like filtering out empty usernames), edit the board layout (widen panels, rearrange sections, group things how you think about them), and add or remove whatever you want.

Tips for getting a great board

Be specific about your service. Tell it the service name, point it to the code, and point it at the right Honeycomb environment.

Tell it what matters. SLOs? Business metrics? Dependency performance? Users? Say so upfront, or ask for additions after the first draft.

Get critical. Tell it, "don't create the board until I tell you." It can show you (and link you to) the queries it's going to include, and you can critique them. Of course, you can change anything on the board afterward, but it's fun when it gets it right the first time.

Get really specific with a skill. Take the skill I used in the video, copy it, and edit it with your own opinions and aesthetics.

When it's wrong, try again. Not satisfied? Start over. Delete the board it created, clear its memory, change your prompt or skill, and have it try again. The sunk cost is small when an agent is doing the work.

Try it yourself

This works with any AI agent that supports MCP: Claude Code, Cursor, and more. You need a Honeycomb account with Honeycomb Intelligence enabled (free accounts work). If your software isn't sending data yet, tell your agent to instrument with OpenTelemetry and send to Honeycomb. Point it to docs.honeycomb.io for details.

You don't have to be a dashboarding expert, or spend a lot of time on it, or settle for a generic dashboard.

Get the nice entry point you want, and then drill into any graph, getting down to the individual events and whole traces. That's the Honeycomb magic.