10 monitoring MCP servers for Claude Code, compared (2026)

Already on Datadog, Sentry or Grafana? Use that vendor's official MCP server: each one adds to Claude Code with a single command or an official plugin. If errors are all you need, Sentry's free plan includes MCP access. If you have no monitoring yet, pick one tool that covers errors, logs and uptime, such as Better Stack or Fixter.

We checked every server below against the vendor's own docs, repos and pricing pages on October 16, 2026, and cite them inline. Fixter is our product. It is in the list, not at the top: the three most searched servers come first, then the rest in alphabetical order.

The ten servers at a glance

An MCP server is how Claude Code talks to another tool. A monitoring MCP server lets it read your production data: search logs, open an error with its stack trace, follow a slow request through its trace. If the term is new, start with what an MCP server is.

Lowest list price in each vendor's currency on October 16, 2026; Datadog, Sentry and Better Stack billed yearly. "Hosted" means the vendor runs the MCP server and you connect to a URL. Click a tool for details, monthly prices and sources.

Each server, with its install line

Lines that start with /plugin run inside Claude Code. Lines that start with claude run in your terminal. Every hosted server here signs you in through the browser on first use, except Peeko, which takes an API key.

Datadog MCP server

Datadog's MCP server is remote, run by Datadog, and generally available since March 2026.¹ The default toolset reads logs, metrics, traces and spans, monitors, incidents, dashboards, hosts and services. Error tracking, synthetic tests and several more are extra toolsets you switch on with a toolsets parameter on the URL.² Datadog states fair-use limits of 50 requests per 10 seconds and 100,000 tool calls a month.³ The Claude Code plugin is in preview and needs Claude Code 2.1.30 or later.⁴

# official plugin (preview), inside Claude Code
/plugin install datadog@claude-plugins-official

# or the server alone; the endpoint depends on your Datadog site
claude mcp add --transport http datadog-mcp <your Datadog MCP endpoint>

For a small team the catch is the bill, not the MCP server. Datadog prices each product on its own: infrastructure per host, logs per GB ingested and per million events indexed, APM per host. The free tier is infrastructure only, up to 5 hosts, and Infrastructure Pro starts at $15 per host a month billed yearly.⁵ We work through what that adds up to in Datadog alternatives for small teams.

Sentry MCP

Sentry runs a hosted MCP server at mcp.sentry.dev, and all connections sign in with OAuth.⁶ It reads issues with their stack traces and breadcrumbs, traces and spans, logs, replays and uptime monitors, and it can ask Seer, Sentry's AI, to analyse an issue.⁷ Seer is a paid add-on at $40 per active contributor a month.⁹ A local version (npx @sentry/mcp-server) also works with self-hosted Sentry.⁷ The official plugin bundles the hosted server with Sentry's skills for SDK setup and debugging.⁸

# official plugin, inside Claude Code
/plugins install sentry@claude-plugins-official

# or the server alone; the org and project part of the path is optional
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}

MCP access is on the free Developer plan: one user, 5,000 errors, 5 GB of logs and one uptime monitor a month. Team starts at $26 a month billed yearly, with unlimited users.¹⁰ Sentry is error tracking first; for the wider picture see Sentry alternatives.

Grafana MCP

Grafana has two official servers. Grafana Cloud runs a hosted one at mcp.grafana.com, generally available since July 2026, which signs in with OAuth and acts with your Grafana role.¹¹,¹² For self-hosted Grafana, the open-source mcp-grafana runs on your machine or cluster (a binary, Docker, uvx or Helm) with a service account token.¹³ Both read through your data sources: Prometheus metrics, Loki logs, Tempo traces, plus dashboards, alerting and incidents. Neither lists uptime checks as a tool.¹³

# Grafana Cloud, official plugin, inside Claude Code
/plugin marketplace add grafana/ai-marketplace
/plugin install grafana-cloud-mcp@grafana-ai-marketplace

# Grafana Cloud, the server alone
claude mcp add grafana --transport http https://mcp.grafana.com/mcp

# self-hosted Grafana, with mcp-grafana installed locally
claude mcp add grafana -- mcp-grafana

On Grafana Cloud the hosted server counts toward Grafana Assistant: each person who connects is an active AI user, though read-only tool calls use no tokens.¹¹ Grafana Cloud's free tier has 50 GB of logs, 50 GB of traces, 10,000 metric series and three users, including three active AI users. Pro starts at $19 a month plus usage, and Assistant on Pro is $20 per active AI user.¹⁴ Grafana is the pick when you already run it. Starting from zero, you assemble the pieces yourself, and the agent can only query what you have wired into Grafana.

Axiom MCP

Axiom's server is hosted at mcp.axiom.co, with a local Go version if you would rather run it yourself.¹⁵ The agent queries your events and logs in APL, Axiom's query language, queries metrics, and manages dashboards and monitors. There are no uptime checks.¹⁶ The hosted server runs in the US, so query results pass through US infrastructure.¹⁵ MCP is included on every plan; agent queries count toward query compute, which you cap per role.¹⁵,¹⁷ The free Personal plan has 500 GB a month of data loading and 25 GB of storage; Axiom Cloud is $25 a month plus usage.¹⁷

claude mcp add --transport http axiom https://mcp.axiom.co/mcp
# then run /mcp inside Claude Code to sign in

Better Stack MCP

Better Stack's server is hosted at mcp.betterstack.com and signs in with OAuth or an API token.¹⁸ It reads logs, spans, metrics and errors through SQL queries, and it reaches uptime monitors, heartbeats, incidents, on-call schedules and status pages.¹⁸ The official plugin adds skills for investigating incidents, querying logs and errors, and managing uptime monitors.¹⁹ The pricing page lists the MCP server as a feature with no separate price.²⁰

# official plugin, inside Claude Code
/plugin marketplace add BetterStackHQ/claude-plugin
/plugin install betterstack@betterstack

# or the server alone
claude mcp add --transport http betterstack https://mcp.betterstack.com

Better Stack is the closest to Fixter in scope: errors, logs and uptime in one product. It is a full web app with MCP as one way in; Fixter is built to be set up and run from your coding agent. Free covers 10 monitors and heartbeats, 100,000 exceptions, 3 GB of logs and 3 GB of traces; paid starts at $29 per responder seat a month billed yearly, or $34 monthly.²⁰

Dash0 MCP

Dash0 is an OpenTelemetry-native platform with a hosted MCP server; you copy its endpoint from your Dash0 account.²¹,²² The agent reads metrics, logs and traces, manages dashboards and alert rules, and can run Agent0, Dash0's investigation agent. Plain telemetry queries over MCP are free; an Agent0 investigation costs about one to two credits, at $0.60 a credit.²¹,²³

# the endpoint is shown in your Dash0 account
claude mcp add dash0 <your Dash0 MCP endpoint> --transport http

There is no free tier, only a 14 day trial. After that, billing is usage based: $0.06 per million spans or log records ingested, plus $0.54 per million stored.²³

Fixter MCP (ours)

Fixter is ours, so weigh this section with that in mind. Fixter finds and fixes the issues in your app in production, from your coding agent: errors, failing infrastructure, slow requests and downtime. It finds the root cause and fixes it: in your agent over MCP, or as a pull request ready for your review. Nothing to learn, built for small teams.

The difference from the rest of this list: Fixter is not an MCP add-on to a platform you already run. It is a monitoring tool you set up from Claude Code when you have none. Your app sends logs, traces and metrics over standard OpenTelemetry, with no Fixter SDK.²⁴ Over MCP the agent searches logs, walks traces, queries metrics, runs read-only SQL, works through tracked issues, manages alert rules and uptime probes, and on paid plans picks up investigations to continue in your editor.²⁵,²⁶ Alerts reach you by email or Slack.

# paste this into Claude Code and it does the whole setup:
Read fixter.dev/llms.txt and set up Fixter.

# or connect the server alone, in a terminal
claude mcp add fixter --transport http https://mcp.fixter.dev
# then run /mcp inside Claude Code to sign in

Querying over MCP is free on every plan. Free plan: 10 GB a month and 1,000 uptime checks. Paid from €20 a month.²⁷ Where it falls short: there is no self-hosted option, and it is a younger, smaller product than Datadog, Sentry or Grafana. If your team already lives in one of those, adding its MCP server is less work than moving. See what Fixter does.

Honeycomb MCP

Honeycomb's hosted MCP server has been generally available since September 2025, at no extra charge on all accounts.²⁸ It runs queries over your events and traces, including BubbleUp, Honeycomb's way of showing what is different about slow or failing requests, and it manages boards, triggers and SLOs. There are no uptime checks.²⁹ Honeycomb Intelligence has to be on for your team.³⁰

# official plugin, in a terminal
claude plugin marketplace add honeycombio/agent-skill
claude plugin install honeycomb

# or the server alone (EU: https://mcp.eu1.honeycomb.io/mcp)
claude mcp add honeycomb --transport http https://mcp.honeycomb.io/mcp

Free covers 20 million events a month with MCP included; Pro starts at $150 a month.³¹

New Relic MCP

New Relic's hosted server, now called New Relic Ground Truth, became generally available in September 2026 and signs in with OAuth only.³² It runs NRQL queries and reads error groups, logs, distributed traces, synthetic monitor results, alerts and incidents; trace search is still in preview.³³ Unlike most servers here, its tool calls are metered: most count as Core Compute Units against your allocation.³³ OAuth needs Claude Code 2.1.186 or later.³⁴

# US region; EU and JP have their own URLs
claude mcp add --transport http --scope user newrelic https://mcp.newrelic.com/mcp/
claude mcp login newrelic

The free tier is 100 GB of ingest a month and one full user. Standard is $10 for the first full user and $99 for each one after.³⁵

Peeko

Peeko is website monitoring for Claude and Codex: uptime, API, SSL, domain, cron and heartbeat checks, created and read through a hosted MCP server.³⁶ It does not read your app's errors, logs or traces, so it is an uptime tool, not error tracking. Free is 5 monitors; Pro is $7 a month for 50.³⁶

# the key comes from your Peeko workspace
claude mcp add peeko --transport http https://mcp.peeko.dev/pk_YOUR_KEY

That line is from Peeko's homepage. Its MCP docs page shows a different form, https://peeko.dev/mcp with the key sent as a bearer header, so check both if one fails.³⁷

Three things to check before you connect one

• What it can change. Most of these servers can write as well as read: create monitors, alert rules, dashboards. They act with your permissions. Datadog rejects write calls from a read-only user,³ Honeycomb API keys need Write permission for write tools,³⁰ and Grafana Cloud asks for a separate write scope.¹¹ If you only want Claude reading, give it a read-only role or token.
• Where results go. Query results pass through the vendor's server and into your Claude session. Axiom says its hosted server routes results through the US.¹⁵
• What agent queries cost. An agent asks more questions than a person does. Check New Relic's compute metering,³³ Grafana Cloud's active AI user count,¹¹ Axiom's query compute¹⁷ and Datadog's fair-use limits³ before you let it loose.

Which monitoring MCP server should you use?

The MCP server follows the monitoring tool, not the other way round. Claude can only query data that is already there, so the real choice is where your production data lives.

• Already on Datadog. Install the Datadog plugin and turn on the error-tracking toolset if you use Datadog for errors. It reads what you already pay for. Do not start on Datadog just to get its MCP server: you pay per product, per host.
• Already on Sentry. Install the Sentry plugin. Errors are where it is strongest, and it works on the free plan. If you need more uptime checks than the one free monitor, look at the Team plan or a separate uptime tool.
• Self-hosting Grafana. Use the open-source mcp-grafana next to your Grafana, with a service account token. The hosted server at mcp.grafana.com is for Grafana Cloud only.¹¹
• No monitoring yet. You are choosing a monitoring tool, not an MCP server. Pick one product that covers errors, logs and uptime and includes MCP on its free plan. Sentry if errors come first. Better Stack if you want uptime, incidents and on-call in a web app. Fixter if you want Claude Code to set it up and run it, with nothing new to learn. If you only need to know when your site is down, Peeko is enough.
• Other cases. Honeycomb if you debug mostly through traces. Axiom if you keep a lot of logs and want the storage cheap.

Whichever you pick, the first thing to try is the same: ask Claude about a real error from last week. More on that in how to connect Claude Code to your logs.

Frequently asked questions

How do I give Claude Code access to my production logs?

Connect your monitoring tool's MCP server. In a terminal, run claude mcp add with the server's URL, for example claude mcp add --transport http sentry https://mcp.sentry.dev/mcp, then run /mcp inside Claude Code to sign in. Claude can then search your logs, errors and traces with your permissions. Datadog, Sentry, Grafana, New Relic, Honeycomb, Axiom, Better Stack, Dash0 and Fixter all have one.

Datadog MCP or Sentry MCP for Claude Code?

Use the one for the tool your data is already in. Sentry's is the stronger pick for errors: issues with stack traces and breadcrumbs, plus traces, logs and uptime monitors, and MCP access is on its free plan. Datadog's covers more ground (logs, metrics, traces, monitors, incidents, with error tracking and synthetics as extra toolsets), but you need to be paying for Datadog for it to have anything to read.

Is the Datadog MCP server free?

Datadog's pricing page lists the MCP server without a separate price, and Datadog states fair-use limits of 50 requests per 10 seconds and 100,000 tool calls a month. What you pay for is the Datadog products whose data it reads, such as infrastructure per host and logs per GB.

Does Sentry MCP work on the free plan?

Yes. Sentry's free Developer plan lists MCP access. The plan is one user, 5,000 errors, 5 GB of logs, 5 million spans and one uptime monitor a month. Seer, Sentry's AI analysis, is a paid add-on.

Which monitoring MCP servers can I run myself?

Grafana's open-source mcp-grafana runs on your machine or cluster and works with self-hosted Grafana. Sentry's server also runs locally with npx and supports self-hosted Sentry. Axiom has a local Go version, though your data still lives in Axiom's cloud. The others are hosted by the vendor only.

Do monitoring MCP servers cost extra?

Mostly not. Sentry, Honeycomb, Axiom and Fixter include MCP on their free plans, and Better Stack and Datadog list it without a separate price. The exceptions: New Relic meters most MCP tool calls as compute units, Grafana Cloud counts each person who connects as a Grafana Assistant user, and Dash0 charges credits when you run its investigation agent.