Honeycomb MCP Server Cursor IDE Setup 2026

Honeycomb runs its MCP server as a hosted endpoint — https://mcp.honeycomb.io/mcp — so there's nothing to install or run locally. Add the URL to Cursor's config, authorize with either OAuth or an API key, and your AI can pull traces, run queries, and inspect triggers and SLOs against real production telemetry while you're debugging in the editor instead of tab-switching to the Honeycomb UI.

Honeycomb previously had a self-hosted MCP server option; that path is deprecated in favor of this hosted one. If you're following an older tutorial that has you cloning a repo and running a local process, stop — that's the retired approach, and the hosted endpoint below is both simpler and the one Honeycomb actually maintains now.

Prerequisites

Step 1: Add the Endpoint

{
  "mcpServers": {
    "honeycomb": {
      "type": "http",
      "url": "https://mcp.honeycomb.io/mcp"
    }
  }
}

If your Honeycomb data lives in the EU instance, use the EU endpoint instead: https://mcp.eu1.honeycomb.io/mcp. Pointing a US client config at the wrong region's endpoint is a common first-connection mistake — check which Honeycomb instance your team actually uses before assuming the default.

Step 2: Authorize

Restart Cursor. On first use, you'll choose between OAuth consent (opens a browser to log into Honeycomb and pick an environment and access level) or an API key, depending on which your client supports and your team prefers. Either path lets you explicitly grant the mcp:write scope if you want the AI to create or modify boards, triggers, SLOs, or notification recipients — leave it out if you only want read access, since write tools check for that scope specifically and fail without it.

Step 3: Verify

Query Honeycomb for the p95 latency of the checkout service over
the last hour, and pull any traces above 2 seconds.

What You Can Do With Honeycomb MCP in Cursor

Practical Workflows

Debugging a latency spike mid-coding-session

Something in the checkout flow got slower after my last deploy.
Query Honeycomb for p50/p95/p99 latency on the checkout service
for the last 2 hours, and pull the slowest 5 traces.

Cross-referencing production behavior with the code you're editing

Here's the function I'm about to refactor [paste code]. Pull recent
Honeycomb traces for this service and tell me if there's an existing
error pattern in production I should account for before I change this.

Setting up monitoring for something you just shipped

Create a Honeycomb trigger that alerts if error rate on the new
webhook endpoint exceeds 2% over a 10-minute window.

This one needs the mcp:write scope granted — without it, the request fails at the permission check rather than partially succeeding.

When This Setup Breaks

Write requests fail even though the connection works for queries. This is the scope gotcha, not a bug: read access and write access are separate grants. If board creation, trigger management, or SLO changes fail while querying works fine, go back to your OAuth consent or API key configuration and confirm mcp:write was actually granted — it's opt-in, not bundled with basic access.

Connects but returns no data, or data from the wrong team. Check the endpoint region first (US mcp.honeycomb.io vs. EU mcp.eu1.honeycomb.io) — a connection to the wrong region's endpoint succeeds at the network level but simply has no visibility into data stored in the other region.

Following an old guide that has you running a local server binary. That's the deprecated self-hosted path. Replace it with the hosted https://mcp.honeycomb.io/mcp endpoint above — there's no local process to keep running or update anymore.

When NOT to use this MCP server: if your team's Honeycomb access is tightly compartmentalized by environment and you don't want an AI client holding broad query access across environments, scope the OAuth consent or API key to a single environment rather than granting org-wide access, and treat mcp:write as something to grant per-task rather than permanently.

Frequently Asked Questions

Q: Is Honeycomb's MCP server official, or a community integration?
A: Official — it's hosted and maintained by Honeycomb at mcp.honeycomb.io, documented directly in Honeycomb's own docs, not a third-party wrapper.

Q: Do I need to run anything locally, like a Docker container or npx process?
A: No. It's a vendor-hosted remote endpoint reached over HTTP — you point your MCP client at the URL and authorize, with nothing installed or running on your machine. Honeycomb's earlier self-hosted server option is now deprecated in favor of this.

Q: What's the difference between the OAuth and API key auth options?
A: OAuth opens a browser flow to log into Honeycomb and pick an environment and access level interactively; an API key is a static credential configured ahead of time. Both can be scoped to include or exclude the mcp:write permission — the choice between them is more about your team's credential-management preference than a capability difference.

Q: Why can I query traces but not create a trigger?
A: Write operations — creating boards, triggers, SLOs, or notification recipients — require the mcp:write scope explicitly, which is separate from basic read/query access. If write calls fail, that scope likely wasn't granted during OAuth consent or API key setup.

Q: My team is on the EU instance — does the standard setup work?
A: Not with the default endpoint. Use https://mcp.eu1.honeycomb.io/mcp instead of the US mcp.honeycomb.io endpoint; pointing at the wrong region connects successfully but won't see your team's data.

Related Guides

---