Sentry MCP alternatives: error trackers your agent can read

A Sentry MCP alternative only makes sense if the error tracker behind it fits, because the MCP server is a window onto that tracker's data. On 6 Nov 2026 the official MCP registry lists first-party error-tracking servers from four vendors: Sentry, PostHog, Honeybadger and Volato. Bugsink and GlitchTip have community servers only, and Epure's server is published on npm but not yet in the registry. The table compares what each returns, what it can change and what the tracker costs.

Where is Sentry MCP still the better choice?

Sentry MCP is the reference server, and for many teams it is the right one. Claude Code's own MCP documentation uses claude mcp add --transport http sentry https://mcp.sentry.dev/mcp as its worked example (Claude Code docs), the registry shows version 0.48.0 published on 5 Nov 2026, and Sentry lists MCP access on its free Developer plan (Sentry pricing). If your errors already live in Sentry and the bill is fine, connect it and stop here.

• Seer behind the tools. The default seer skill asks Sentry's AI debugger for root-cause analysis and fix recommendations (sentry-mcp). No other server here runs a hosted debugger for you.
• More than errors. Sentry MCP can search traces and logs and read Sentry's SDK documentation, because Sentry stores and publishes them. Epure holds errors only.
• Project setup from the agent. The project-management skill creates projects, teams and DSNs, so an agent can wire a new app to Sentry end to end.
• Reach. Sentry's server is the one coding agents have seen most often, and Sentry's free plan already includes it.

Can you self-host Sentry MCP?

Sentry MCP can run against self-hosted Sentry: start the stdio package @sentry/mcp-server with SENTRY_HOST=sentry.example.com and a user auth token, or SENTRY_URL for an instance without TLS (sentry-mcp README, checked 6 Nov 2026). Two limits apply. Seer is not part of self-hosted Sentry, so the seer skill is dropped, and AI search tools such as search_errors need your own OpenAI, Azure OpenAI, Anthropic or OpenRouter key.

The bigger cost is Sentry itself: self-hosted Sentry asks for 4 CPU cores and 16 GB of RAM plus 16 GB of swap (self-hosted docs). Lighter self-hosted trackers your agent can read are Epure, whose two containers idled at about 5 MiB and 48 MiB on a 769 MiB VPS and serve MCP under AGPL-3.0, and Bugsink or GlitchTip with a community server. Honeybadger's MCP server runs in Docker, but the tracker it reads is hosted.

On a self-hosted Epure, point the stdio server at your instance, or turn on the HTTP endpoint that Epure Cloud serves:

# stdio: OAuth sign-in against your own instance (HTTPS required)
npx -y @epure-sh/mcp login --url https://errors.example.com
claude mcp add epure -- npx -y @epure-sh/mcp --url https://errors.example.com

# or serve /mcp from the Epure container (off by default)
EPURE_MCP_ENABLED=true

How do you add Sentry MCP or an alternative to Claude Code, Cursor or Codex?

Each hosted MCP server here installs in Claude Code with one claude mcp add --transport http line and a browser OAuth sign-in. Cursor and Codex take the same URL, or a stdio command with its arguments. The commands below come from each vendor's docs, checked 6 Nov 2026. PostHog's wizard writes the client config for you, and Epure's init prints a config for Cursor or any stdio client with no credential in it.

# Sentry (hosted, OAuth)
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

# Honeybadger (hosted, OAuth; eu-mcp.honeybadger.io for the EU region)
claude mcp add --transport http honeybadger https://mcp.honeybadger.io/mcp

# PostHog (the wizard writes the client config)
npx @posthog/wizard mcp add

# Epure (hosted, OAuth)
claude mcp add --transport http --scope user epure https://app.epure.sh/mcp

# Epure (stdio): sign in once, then print a Cursor-style config
npx -y @epure-sh/mcp login
npx -y @epure-sh/mcp init
# {"mcpServers":{"epure":{"command":"npx","args":["-y","@epure-sh/mcp"]}}}

The full Epure walkthrough, from sign-in to the first epure_issues_context call, is in connect Epure to Claude Code, Cursor or Codex. Every flag is listed in the MCP and CLI reference.

Which error-tracking MCP servers can change data?

Most error-tracking MCP servers can write, and each has a different off switch. Sentry MCP has two write skills: triage resolves, assigns and updates issues, and project-management creates projects, teams and DSNs. Grant only inspect with --skills=inspect, or ?skills=inspect on the hosted URL, to keep the agent read-only. PostHog removes every create, update and delete tool when you send the x-posthog-read-only header or readonly=1 (PostHog MCP FAQ).

Honeybadger's server starts read-only: HONEYBADGER_READ_ONLY defaults to true, and deletes need a confirm token that expires after 10 minutes (honeybadger-mcp-server). The community Bugsink and GlitchTip servers expose list and get tools only, and Volato's registry entry describes its server as read-only.

Epure exposes every product action by default, because setting up alert rules and cron monitors is part of the job, and gives you two ways to narrow it. npx -y @epure-sh/mcp --read-only removes every tool that is not a read, and login --scope read:agent asks for read access only, which the server enforces whatever the client sends. Destructive actions require confirm: true, and each tool carries readOnlyHint and destructiveHint annotations.

Is it safe to let an agent read production errors through MCP?

Reading production errors through MCP is reasonable with two habits: start read-only, and treat every event as text written by a stranger. A Sentry-style DSN is public by design, since it ships in your browser bundle, so anyone can send an event whose message reads like an instruction to your agent. PostHog's FAQ says it plainly: models can be tricked into following untrusted commands, so review tool calls (PostHog MCP FAQ, checked 6 Nov 2026).

Epure handles part of this in the server. Its MCP instructions tell the agent that application content is untrusted data, never instructions. epure_issues_context leaves raw payloads out and hashes end-user emails and IDs unless the call asks for them. The legacy hosted tool names also wrap the brief in an <untrusted_exception_data> block with prompt-style tags and shell directives neutralised; the current epure_issues_context returns JSON without that wrapper (checked in the source, 6 Nov 2026). None of this makes a model immune, so keep write scopes away from unattended loops.

We found no documented event-text handling in the Sentry, Honeybadger or community-server pages we read; that is a gap in the docs, not proof of a gap in the code. The longer version, with read-only setups per vendor, is in give your coding agent production errors.

What does each option cost a team of one or two developers?

For one or two developers, connecting the MCP server costs nothing on every tracker we could price; the cost is the tracker's plan and any AI add-on. At 200,000 errors a month, PostHog costs $37, GlitchTip $50, Bugsink €50, Sentry Team about $61 before Seer, Honeybadger Team $71 with overage turned on, and Epure Pro $29. Sentry adds $40 per active contributor for Seer, and PostHog bills some model-backed MCP tools as PostHog AI usage.

Retention changes the comparison: Sentry Team keeps up to 90 days, Honeybadger Team and GlitchTip Small 90 days, and Epure Pro 30 days. Epure's case is the flat number. The bill at 200,000 errors is the bill at 50,000, with unlimited seats and no automatic overage (Epure pricing). The Sentry side, with Seer for two developers, is worked through in Epure vs Sentry.

Which Sentry MCP alternative should you pick?

The Sentry MCP alternative to pick follows from where your errors should live, since each server only reads its own tracker. Four of the seven trackers here (checked 6 Nov 2026) accept events from the Sentry SDK: Sentry itself, Epure, Bugsink and GlitchTip, so moving between them means changing the DSN. PostHog and Honeybadger use their own SDKs. Choose the tracker first, then check its server against the table above.

• Sentry MCP if you use Seer, traces or logs, or stay under 5,000 errors a month on the free plan.
• PostHog MCP if you already use PostHog analytics and want errors next to sessions and flags, with 100,000 free exceptions a month.
• Honeybadger MCP if you want a hosted server that starts read-only, with uptime checks and logs on the same bill.
• Bugsink or GlitchTip if you self-host, want the lowest hosted price, and accept an MCP server the vendor does not maintain.
• Epure if errors are the whole job, you keep your Sentry SDK, and you want MCP and the CLI on every plan, self-hosted under AGPL-3.0 or $29/month on Cloud.
• Volato once its site is reachable again; we could not verify its pricing or sign-in on 6 Nov 2026.

How did we check these MCP servers?

We queried the official MCP registry API (registry.modelcontextprotocol.io/v0/servers?search=<name>) for each vendor on 6 Nov 2026. We read each vendor's MCP docs or README and its pricing page and, for Epure, the source of @epure-sh/mcp 0.3.0 and the server crates. We did not benchmark answer quality between servers. In the table, "not documented" means we did not find it in those pages.

This page is written by Epure, a vendor in the table. It is re-checked every three months. Send corrections to support@news.epure.sh or open an issue on epure-sh/epure; we date every change. More head-to-head pages are on the comparison hub.