ObservabilityMCP servermeasured 19 Aug 2026

Sentry

Sentry's MCP server, published by Sentry as @sentry/mcp-server, gives an AI agent access to your Sentry organization: issues, events, projects, releases, and its Seer analysis. It ships as bundled skills named inspect, seer, docs, triage and project-management, and talks to sentry.io or a self-hosted host over stdio.

Sentry · measured by Keyoor

BootDID NOT BOOTWe started it and it never completed initialize: exited with code 1 before initialize completed

Tools0tools/list payload 0 bytes

Context tax~0estimated tokens for the whole tools/list, chars/4 estimate

Resources / prompts— / —counts returned by resources/list and prompts/list

Measured19 Aug 2026one run of the harness, clean environment, no credentials · how we test

How we started it

npx -y @sentry/mcp-server

Verdict

We started it with a clean environment and no credentials, and it never reached initialize: it printed its usage text to stderr and exited with code 1. That is defensible behaviour for a server whose entire job is reading your error data, but it means the tool list is unmeasurable here and you should only wire it up once you have a Sentry auth token in hand. Use it if you already live in Sentry; do not expect it to be a drop-in that boots and shows you what it can do first.

Use it if

You already triage errors in Sentry daily and want an agent pulling the stack trace instead of you tab-switching

You have a scoped Sentry auth token you can put in SENTRY_ACCESS_TOKEN and rotate later

You run self-hosted Sentry and need the --host flag to point an agent at an internal instance

You want the Seer analysis surfaced inside your editor agent and are willing to hand over a model API key for it

Skip it if

You wanted to evaluate the tool list before committing credentials, because without a token there is nothing to list

Your policy forbids giving an MCP process both your production error data and an OpenAI or Anthropic key at once

You only need occasional error lookups and the Sentry web UI already does that in two clicks

You are running agents in CI where the device-code OAuth flow has nobody to approve it

You are on self-hosted Sentry over plain HTTP and cannot justify the --insecure-http flag to whoever owns security

The bill for your context window

There is no Context Tax to report for this one, because the server exited before tools/list and our measurement records zero tokens across zero tools. Read that as unmeasured, not as free: once authenticated it exposes five skill bundles, and skill-bundle servers are usually the heavy end of the directory. If you install it, run a tools/list yourself with the token in place and trim with --disable-skills before deciding what it costs your window.

Setup reality

Auth first, everything else second. The stderr help text lists SENTRY_ACCESS_TOKEN (or the equivalent --access-token flag), SENTRY_CLIENT_ID to override the OAuth client for the device-code flow, and OPENAI_API_KEY, ANTHROPIC_API_KEY or OPENROUTER_API_KEY for the AI-powered search tools, with EMBEDDED_AGENT_PROVIDER choosing between openai, azure-openai, anthropic and openrouter. Without a token it falls back to device-code auth against sentry.io only, which is interactive and therefore useless in a headless probe like ours. Self-hosted users add --host, plus --insecure-http for non-TLS internal hosts. MCP_DISABLE_SKILLS or --disable-skills trims the skill set, which is the knob to reach for if you do not want Seer touching things.

SENTRY_ACCESS_TOKENSENTRY_CLIENT_IDOPENAI_API_KEYANTHROPIC_API_KEYOPENROUTER_API_KEY

Questions people ask

Why did the Sentry MCP server fail to boot in your test?

We ran it as an unprivileged user with a clean environment and no secrets, the way we run every server here. With no SENTRY_ACCESS_TOKEN it printed its usage and environment-variable help to stderr and exited with code 1, before the initialize handshake completed.

Can I run it without giving it an OpenAI or Anthropic key?

Yes. The help text presents those keys as being for the AI-powered search tools, not for the server as a whole, and EMBEDDED_AGENT_PROVIDER only picks between them. A Sentry token alone should get you the inspect and triage side.

Does it work with self-hosted Sentry?

Yes, via --host, and the usage examples show both a hostname and a host:port form. Device-code auth is documented as sentry.io only, so self-hosted means passing an access token directly. Non-TLS internal hosts need --insecure-http.

How this page is made: the server is spawned as an unprivileged user with a clean environment and no credentials, then asked for its tools, resources and prompts over stdio. Token figures are estimates at four characters per token, not a tokenizer count. One run, one machine. Corrections: contact the desk.

← All measured MCP servers