What it is
Sentry MCP connects coding assistants such as Claude Code, Cursor and VS Code to your Sentry account. Sentry says it is built for human-in-the-loop debugging rather than as a general Sentry admin tool, so the tools focus on finding issues, reading errors, traces and logs, and asking Seer, Sentry's AI debugger, for a root cause.
The main way to use it is the hosted server at https://mcp.sentry.dev/mcp with OAuth, optionally scoped to one organisation or project in the URL. A stdio version (npx @sentry/mcp-server) exists for self-hosted Sentry and needs a user auth token. There is also a Claude Code plugin that adds a Sentry subagent. The source lives in Sentry's getsentry/toolkit repository alongside the Sentry CLI; the latest MCP release is 0.42.0, from 25 September 2026.
Install
https://mcp.sentry.dev/mcpOur review
If your team already sends errors to Sentry, this server closes the loop between the error and the code. An agent can pull the stack trace, breadcrumbs and trace for an issue, then fix the code in the same session. The hosted endpoint uses OAuth, and the URL can be scoped to one organisation or project, which hides discovery tools and keeps every call pointed at the right place.
The tool catalogue is larger than it first looks, at about 70 tool files in the source, but it is grouped into five skills. Only inspect and seer are on by default; triage (resolve, assign, update issues), docs and project management are opt-in, and ?skills= or --disable-skills narrows them further. That is the right default for a server whose write tools can create projects, change alert rules and delete monitors.
There are catches for anyone off the hosted path. The natural-language search tools (search_errors, search_issues, search_traces, search_logs) translate questions into Sentry query syntax with an LLM, so a self-run stdio server needs an OpenAI, Anthropic, Azure OpenAI or OpenRouter key or those tools disappear. The stdio token needs project:write, team:write and event:write scopes, broader than a read-only debugging job requires. And Seer is not available on self-hosted Sentry, so that skill drops out there.
The data deserves a second thought. Error messages, breadcrumbs and user-feedback reports often contain text your users typed, which means the agent reads attacker-controllable content through this server, much as it would through a GitHub issues server. Sentry publishes how its rate limits work, per caller and per endpoint, but not the numbers, and warns that polling trips them quickly.
Where it earns its place
A backend engineer at a 15-person SaaS company on Sentry Team
Fix the top new production error without opening the Sentry UI
Add https://mcp.sentry.dev/mcp/<org>/<project> to Claude Code, ask for the latest unresolved issue, and let the agent read the stack trace and breadcrumbs before editing the code. The server is free; the Team plan is $26 a month billed annually.
An engineering manager with 10 developers opening pull requests
Get a root cause and suggested fix for regressions without a person triaging each one
Enable the seer skill so the agent can request Seer's analysis for an issue. Seer bills $40 per active contributor (two or more PRs a month in a connected repo), so ten active developers add $400 a month.
A platform engineer running self-hosted Sentry behind a VPN
Let an internal coding agent read errors without sending data to sentry.io
Run npx @sentry/mcp-server --host=sentry.internal with a user token and EMBEDDED_AGENT_PROVIDER set; disable project-management so the write scopes the token needs are not exposed as tools. Seer is not available on self-hosted.
Before you commit
- Setup
- claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/<org>/<project>, then sign in with OAuth on first use; the same URL works in Cursor and VS Code. Self-hosted users run npx @sentry/mcp-server@latest --access-token=... --host=..., and a first query takes minutes.
- Integrations
- About 70 tools across five skills: inspect (get_issue_details, get_event_stacktrace, get_trace_details, search_events and others), seer (analyze_issue_with_seer), triage (update_issue, add_issue_note), docs and project management (create_project, alert rules, monitors). Only inspect and seer are on by default.
- Data and privacy
- The server reads whatever your Sentry account can see, including error payloads, logs, replays and attachments. The hosted server is run by Sentry, which already holds the data; Sentry lists SOC 2 and ISO 27001 on every plan. Error messages and user feedback can carry prompt-injection text into the agent.
- Cost at scale
- The server is free. Team is $26 and Business $80 a month billed annually, with 50k errors included; Seer adds $40 per active contributor per month. API limits are per caller and endpoint, and numbers are not published.
- Limits
- Stdio is still described as work in progress; AI search tools vanish without an LLM key; Seer tools are missing on self-hosted. Sentry's pricing table marks API access as Team and above, and does not say whether the hosted MCP server works on the free Developer plan.
What's good
- Hosted with OAuth; URL can be scoped to one org or project
- Write skills (triage, project management) are off by default
- Works with self-hosted Sentry over stdio
- Seer root-cause analysis available as a tool on paid plans
What's not
- AI search tools need your own LLM key when self-run
- Stdio token needs write scopes and is labelled work in progress
- Error and feedback text is untrusted input for the agent
- API rate limits are not published as numbers
Best for
Teams that use Sentry and want coding agents to read real error and trace data while fixing bugs.
Alternatives
Sentry MCP: common questions
- Is Sentry MCP worth it?
- If your errors already land in Sentry, this is the shortest path from a stack trace to a fix, and the project-scoped URL keeps the agent from wandering. We rate it 4.3 out of 5.
- How much does Sentry MCP cost?
- The MCP server is free. Sentry has a free Developer plan for one user; Team is $26/month and Business $80/month billed annually; Seer costs $40 per active contributor per month on Team or Business.
- What are the downsides of Sentry MCP?
- AI search tools need your own LLM key when self-run; stdio token needs write scopes and is labelled work in progress; error and feedback text is untrusted input for the agent; aPI rate limits are not published as numbers.
- Who is Sentry MCP best for?
- Teams that use Sentry and want coding agents to read real error and trace data while fixing bugs.
- How much does Sentry MCP cost for a team or in production?
- The server is free. Team is $26 and Business $80 a month billed annually, with 50k errors included; Seer adds $40 per active contributor per month. API limits are per caller and endpoint, and numbers are not published.
- Is Sentry MCP safe for company data?
- The server reads whatever your Sentry account can see, including error payloads, logs, replays and attachments. The hosted server is run by Sentry, which already holds the data; Sentry lists SOC 2 and ISO 27001 on every plan. Error messages and user feedback can carry prompt-injection text into the agent.
- What are the best Sentry MCP alternatives?
- The closest alternatives we have reviewed are GitHub MCP Server and Grafana MCP Server.
Answers checked Oct 23, 2026.
Make Sentry MCP? Show this rating on your site
