Honeycomb MCP server: 34 rate-limited tools

The only connector in this guide that publishes a per-tool rate limit, what its nine write tools need, and the Honeycomb Intelligence setting a team owner must switch on before any of it works.

Last verified 1 September 2026 · from Honeycomb's own MCP documentation and the archived repository's GitHub record

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Honeycomb MCP server?

It is Honeycomb's server for the Model Context Protocol. Honeycomb stores wide events, so a connected agent can run aggregations and follow a trace without opening the query builder.

The tool list follows how an investigation actually goes. get_workspace_context orients the agent. find_columns locates fields by what they describe rather than by exact name. run_query computes the aggregation. run_bubbleup explains what makes a slow subset different from the baseline.

Two tools are unusual. canvas_agent_invoke starts a Canvas investigation and canvas_agent_poll_response collects the result, so this connector can call Honeycomb's own agent, not only read Honeycomb's data.

Honeycomb has two MCP servers, and only one is live:

A search for the repository name still finds the archived project first. The hosted endpoint is the one Honeycomb documents and the one this page measures.

It is strongest in production observability, reaches agent observability through its own conversation tools, and reaches nothing in code review, CI or cost.

In practice, Honeycomb answers what production did and why one slice differs from another. It can compute an aggregation, follow a trace and run a BubbleUp analysis. It has no view of the code or the pipeline, so a fix still needs a repository connector.

What can an AI agent do with Honeycomb?

get_workspace_context is the documented first call: it returns the team name, the current time and every environment with its dataset count, which is what the other tools need.

What connecting costs before the first question

A context window is the amount of text a model can hold at once. Thirty-four tools cost 3,204 to 4,079 tokens on Claude at connection. Nine of them write, and the split is visible in the tool names rather than hidden behind a flag.

Honeycomb writes long descriptions, several sentences each explaining when an agent should reach for the tool. That is why thirty-four Honeycomb tools cost more than fifty-five tools cost elsewhere.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted from the 34 tool entries in the vendor's own documentation. That documentation is the only public inventory.

The server could not be run for a live tools/list. No input schemas are available, so these figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 are at the low end. Opus 5 and Sonnet 5 are at the high end. The gap between them is about a quarter.

Can Honeycomb watch your own AI agents?

Honeycomb reaches one agent signal, and it is not the one most readers expect. Read from Honeycomb's own tools reference on 24 August 2026.

Honeycomb MCP server limits: where answers come back incomplete

Two limits below are about what a number means; the third is about which server a search finds first.

How to configure Honeycomb MCP for agents

1. 1

   Enable Honeycomb Intelligence first

   The documentation names it as a prerequisite, so nothing connects until a team owner has switched it on.

2. 2

   Prefer OAuth

   Honeycomb's own guidance is to use an API key only when the agent cannot support an interactive login.

3. 3

   Scope an API key narrowly if you need one

   It is a Management API key with the Model Context Protocol and Environments scopes; read is required for every operation, and write only for the nine writing tools.

4. 4

   Pick the right region

   The US endpoint is mcp.honeycomb.io/mcp and the EU endpoint is mcp.eu1.honeycomb.io/mcp. They are separate deployments.

Paste this into your agent’s instructions

Honeycomb samples data and its query tools return sampling metadata that an answer can quietly drop. The block keeps it in the answer.

When you answer using Honeycomb MCP tools, state:
- The environment and dataset the answer covers.
- Whether the result carried sampling metadata, and whether
  sample-rate correction was on.
- Whether you created or updated a Board, Trigger, SLO,
  recipient or Canvas investigation.
Never state that data does not exist. State that no events
matched the query above.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for a dashboard or a scheduled report. Honeycomb's query API is the alternative, and for those it is the better one.

The connector earns its place during an investigation. find_columns matches fields by intent and run_bubbleup explains a difference, which are hard to reproduce with fixed API calls.

Canvas is a third option inside the same connection: canvas_agent_invoke hands the investigation to Honeycomb's own agent rather than driving it tool by tool.

Connect it while investigating. Use the query API for anything scheduled or repeated.

Honeycomb MCP server setup

The server is hosted, so there is nothing to install. Honeycomb Intelligence has to be enabled on the team before any client can connect.

OAuth is the documented route for interactive clients. Only team owners can generate the API key used by unattended agents.

claude mcp add --transport http honeycomb \
  https://mcp.honeycomb.io/mcp

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

[mcp_servers.honeycomb]
url = "https://mcp.honeycomb.io/mcp"

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

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

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

{"servers":{"honeycomb":{"url":"https://mcp.honeycomb.io/mcp"}}}

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

{"mcpServers":{"honeycomb":{"command":"npx",
  "args":["-y","mcp-remote",
    "https://mcp.honeycomb.io/mcp",
    "--header","Authorization: Bearer $HONEYCOMB_API_KEY"],
  "env":{"HONEYCOMB_API_KEY":"<KEY_ID>:<SECRET_KEY>"}}}}

{"mcpServers":{"honeycomb":{
  "url":"https://mcp.eu1.honeycomb.io/mcp"}}}

Read-only access, permissions and security

• Writes are scoped separately

  The nine writing tools require the mcp:write scope, and an API key's environment access must cover any environment the agent writes to.

• Only team owners can mint a key

  Honeycomb documents that restriction, which keeps the unattended route from being self-service.

• Rate limits are published per tool

  get_service_map allows 10 calls a minute and creation tools allow 20, so an agent that loops gets a retry time back rather than a wrong answer.

• Keys should carry nothing else

  Honeycomb's own advice is that a key used for MCP holds no other scopes and is kept out of source control.

Troubleshooting

The client cannot connect at all

Honeycomb Intelligence has to be enabled for the team. It is named as a prerequisite before any client configuration.

A write tool is refused

Write tools require the mcp:write scope. An API key needs the Model Context Protocol scope with write granted, and environment access covering the target environment.

Calls start failing after a burst

Limits are per tool and per minute. get_service_map allows 10, creation tools 20, and most others 50. Rate-limit responses include the time you can retry.

Numbers look wrong on a busy dataset

Sampled data returns sampling metadata with the result, and run_query has a usage mode that disables sample-rate correction. Check which was in effect.

Honeycomb MCP server: Reference

What engineers report

Honeycomb publishes more operational detail about its server than any other vendor here, so the counts below come from its own reference.

Tool inventory, rate limits and scopes read from Honeycomb's own documentation on 24 August 2026. An unauthenticated tools/list against mcp.honeycomb.io/mcp returned 401 on the same date, which is why this page is sourced from documentation rather than from the wire. Repository facts from the GitHub API.

Should you connect an agent to Honeycomb?

Yes. Best for OAuth from an interactive client, with an API key reserved for unattended agents and scoped to nothing else.

Strongest fit

Honeycomb teams who want an agent to investigate production the way the query builder does.

Main advantage

Published per-tool rate limits and scopes, which no other connector in this guide provides.

Main weakness

Nothing works until Honeycomb Intelligence is enabled, and only team owners can mint an API key.

Operational risk

Nine write tools can create Boards, Triggers, objectives and recipients when the write scope is granted.

Frequently asked questions