Honeybadger

AI-Native Application Monitoring & Observability

Your data, your agents

Honeybadger gives your coding agent the data it needs to debug production—and the tools to do it right.

claude code
❯ users are getting 500s on checkout. what's going on?
⏵ honeybadger.list_faults (project: Monolith, status: open)

1 error spiking in the last hour:

ActionView::Template::Error · CheckoutController#create

214 occurrences · 47 users · started 11:02am

⏵ honeybadger.get_fault (id: 132044032) 
app/views/checkout/_summary.html.erb:18
undefined method `total' for nil:NilClass

first seen right after deploy b7e204f — "Refactor cart totals"

Found it. Deploy b7e204f stopped setting @cart in the checkout controller, so the view calls .total on nil.

Fixed and committed. Creating PR...

From on-call to all-clear

Fixing the bug is half of any incident. The other half is keeping your team informed, monitoring the recovery, and compiling the postmortem.

Honeybadger helps you coordinate the response, so you stay on task and your team stays in the loop.

❯ put checkout error rate, p99, and deploys on one dashboard so the team can watch the fix land
⏵ honeybadger.query_insights
⏵ honeybadger.create_dashboard

Done. Error rate, p99, and responses by status, with the deploy list beside them so b7e204f lines up with the spike.

app.honeybadger.io/projects/1234/dashboards/checkout-incident

❯ fix is deployed. are we recovered?
⏵ honeybadger.query_insights
⏵ honeybadger.update_fault (resolved)
⏵ honeybadger.create_fault_comment

Yes. Error rate back to 0.2% since 10:35, p99 back near 400ms. Resolved the error and left a comment linking the deploy, the cause, and the dashboard.

For your everyday work

Fix bugs, answer support questions, set up alerts, and instrument new services, all from the same conversation.

Fix a bug

❯ can you fix this? app.honeybadger.io/projects/1234/faults/98123401
⏵ honeybadger.get_fault
⏵ honeybadger.list_fault_notices

NoMethodError in InvoiceMailerJob: the account was deleted after the invoice was queued. Added a guard and a test on fix/invoice-mailer. Want a PR?

Answer a support ticket

❯ user 4821 says checkout failed this morning. what happened?
⏵ honeybadger.query_insights
⏵ honeybadger.get_fault

Two failed POSTs to /checkout at 10:06 and 10:14, both ActionView::Template::Error from the b7e204f deploy. Their order never saved. Fixed at 10:30, no failures since.

Set up alerts

❯ page us if checkout p99 stays over 2s for 10 minutes
⏵ honeybadger.create_alarm 
filter request.path == "/checkout"
| filter environment == "production"
| stats percentile(99, duration) as p99
| filter p99 > 2000

Alarm created, threshold 2000ms, production only. Routed to the existing PagerDuty integration.

Instrument a new service

❯ instrument this service with wide events. one event per request with everything we'd want when debugging
⏵ read app/ (routes, middleware, jobs)
⏵ honeybadger.query_insights

Added one wide event per request: path, status, duration, user and account, plan, feature flags, and DB time. Events are arriving in Insights. Want a dashboard for p99 and errors by path?

Everything in context

With your errors, your data, and your code in one place, your agent hits the ground running.

  1. 1

    You tell it what's wrong

    Paste a link, share a screenshot, or ask a question.

  2. 2

    Your agent investigates

    It pulls the error, the affected users, and the recent deploys—then traces the cause to your code.

  3. 3

    You steer, it does the legwork

    Ask for the fix, a dashboard for the team, or an alarm so the next one gets caught early.

  4. 4

    Your agent closes the loop

    After the deploy, it verifies the fix, resolves the error, and drafts the postmortem.

Built for your agents

Honeybadger's plugin gives your coding agent everything it needs to fix production. Using a different agent? Each piece works on its own, too.

Skills

Knows the job

Skills teach your agent how to debug errors, add Honeybadger to an app, set up alerts, and investigate incidents.

MCP server

Safe access to production

Connect any MCP client in one step. Start read-only, and allow changes when you trust it.

Docs for agents

Docs your agent can read

Point it at llms.txt, or add .md to any docs URL. It looks things up instead of guessing.

Prefer the shell? The hb CLI can do anything the MCP server can, with JSON output for agents and scripts.

Free with your Honeybadger account

Put your agent on call

Install the plugin, connect your account, and start fixing.

  • Claude Code
  • Codex
  • Amp
  • Pi
  • OpenCode
Get the plugin Start free trial