Question 1

What is Honeycomb MCP?

Accepted Answer

Honeycomb MCP is Honeycomb’s hosted implementation of the Model Context Protocol, an open standard that enables AI agents to connect directly to external data sources such as telemetry data, including traces, metrics, and logs.

With Honeycomb MCP, AI agents can query, analyze, and visualize Honeycomb observability data using natural language or tooling interfaces, allowing teams to integrate observability into IDEs, CLIs, and automated workflows. MCP is designed to reduce friction in AI-assisted investigation, help diagnose issues, and surface insights that would otherwise require manual exploration in the Honeycomb UI.

Question 2

Does Honeycomb work with Claude Code?

Accepted Answer

Yes. Honeycomb MCP supports integration with Claude Code and other MCP-compatible AI tools. Using OAuth or API key authentication, you can configure Claude Code to connect to Honeycomb’s MCP endpoint so that the agent can run queries, fetch traces, and explore observability data directly from within Claude Code. This integration leverages the MCP standard to ensure consistent tooling across AI assistants.

Question 3

Does Honeycomb work with ChatGPT?

Accepted Answer

Yes. AI tools like ChatGPT that support the MCP standard with custom connector or agent support can connect to Honeycomb’s MCP server once configured with the correct endpoint and credentials. Once connected, ChatGPT can use Honeycomb MCP to perform structured observability queries instead of relying solely on static knowledge.

Question 4

How do I set up the Honeycomb MCP server?

Accepted Answer

To set up the Honeycomb MCP server:

Ensure your team has Honeycomb Intelligence enabled.

Configure your preferred AI assistant or agent with the MCP endpoint using OAuth 2.1 (recommended) or an API key.

Select the appropriate Honeycomb MCP endpoint, authenticate via OAuth, and authorize access to your Honeycomb team’s data.

Once connected, your AI agent can immediately begin issuing observability queries using the MCP tools.

Question 5

Can my AI agent query Honeycomb directly?

Accepted Answer

Yes, an AI agent can query Honeycomb directly through the MCP server, which acts as an intermediary between the AI agent and Honeycomb’s observability data.

The AI agent must support MCP and be configured with the Honeycomb MCP endpoint using the appropriate authentication method. Once connected, the agent can run queries, fetch traces, use BubbleUp, and inspect triggers and SLOs, just like you’re interacting with Honeycomb’s APIs natively.

Question 6

How does Honeycomb MCP handle API rate limits?

Accepted Answer

Honeycomb MCP enforces rate limits on tool calls to protect service stability and performance. If an AI agent exceeds these limits, requests may return a rate limit exceeded error, indicating a temporary block until the quota resets.

Honeycomb MCP exposes multiple distinct tools that AI agents can call, with each tool mapping to a specific operation, such as running a query, fetching traces, or exploring schema metadata. Rate limits are applied per tool call rather than per user action in aggregate. Most MCP tools are limited to approximately 50 calls per minute per user, while certain specialized tools, such as get_service_map, have more bounded limits (for example, 10 calls per minute).

To avoid hitting rate limits, design prompts and workflows to minimize unnecessary MCP tool calls, and consider upgrading your plan if your use case needs a sustained higher throughput.

Question 7

What is MCP observability?

Accepted Answer

MCP observability makes AI agent workflows transparent, measurable, and traceable when they’re interacting with your systems via MCP. It involves instrumenting, logging, and monitoring agent-initiated operations such as queries, trigger checks, and anomaly investigations so teams can understand how AI agents use observability data, track performance, and detect patterns or inefficiencies. These concepts expand traditional system observability to include insights into how automated agents interact with telemetry and tooling.

Question 8

What are the main components of MCP observability?

Accepted Answer

MCP observability is composed of several core components that enable AI agents to safely and effectively interact with observability data. At its core are the MCP tools, which perform specific actions like querying metrics, fetching traces, analyzing logs, and inspecting schemas. Supporting these tools are tool invocation logs, which capture when and what agents call specific MCP tools, and telemetry data that captures volume, latency, and error rates. Observability MCPs also include usage analytics to understand which tools and datasets are accessed most frequently, as well as rate limiting and access controls to protect system stability.

Together, these components provide visibility into AI agent behavior, enable debugging of agent workflows, and ensure agents operate against real, trusted system data rather than inferred or hallucinated context.

Question 9

Why do AI agents hallucinate when debugging real systems?

Accepted Answer

AI agents hallucinate when debugging real systems because they operate on probabilistic language models trained on broad datasets. Without access to authoritative telemetry, such as traces, logs, or metrics, agents fill in gaps with plausible but incorrect outputs. This can result in inaccurate code suggestions, incorrect assumptions about system state, or fabricated explanations. Providing verified context and real-time observability data is crucial to reducing these errors.

Question 10

How do I reduce AI hallucinations when coding?

Accepted Answer

There are several proven ways to reduce AI hallucinations when coding and debugging production systems. One of the most effective methods is to provide AI models with accurate, structured input, including code, telemetry data, and up-to-date documentation. When models have access to authoritative, real-world inputs, they are less likely to make incorrect assumptions or generate misleading answers.

You can also employ techniques like retrieval-augmented generation (RAG), ground AI responses in real data sources, while chain-of-thought prompting and validation loops improve logical consistency and accuracy as outputs are confirmed against known system behavior. And lastly, integrating tools that supply real-time observability data can help reduce AI hallucinations. Tools such as Honeycomb MCP allow AI agents to reason about the actual state of your services, rather than relying on incomplete context or guesswork.

Question 11

How do I set up my agents with Honeycomb Agent Skills?

Accepted Answer

Head to the GitHub repository and follow the readme instructions to get started. Don’t forget, you can fork and edit any of our skills to add information, procedures, and context specific to your organization.

Turn your AI agents into observability experts

AI has accelerated the speed of failure

The intuition gap

Enthusiasm vs understanding

Human-speed infrastructure

Observability built for the future of software

Join us for our virtual event with live Q&A

See it in action

Related features

BubbleUp

LLM Observability

Distributed tracing

Want to know more?

Frequently asked questions