Sentry MCP server: how founders with a coding agent connect it to Claude Code, Codex and Cursor
The Sentry MCP server is a hosted endpoint, https://mcp.sentry.dev/mcp, that lets a coding agent such as Claude Code, Codex or Cursor search your Sentry errors, traces, logs and session replays. It can also ask Seer, Sentry's AI debugger, for a root cause. You connect with an OAuth sign-in. Sentry lists no price for the server itself; its free Developer plan includes 50 replays, Team is $26 a month billed annually, and Seer costs extra.
This page is for founders who build with Claude Code, Codex or Cursor and already send errors to Sentry, or are weighing it. We do not have a Sentry account. Everything about Sentry below comes from Sentry's own pages and source code, each marked with the date we read it, and the setup steps are Sentry's, not a run of ours.
Jump to: setup · tool groups · replay tools · access and limits · price · where Sentry wins · where UserProof fits · sources · FAQ
How do you add Sentry's MCP server to a coding agent?
Sentry runs one remote endpoint, https://mcp.sentry.dev/mcp, and every remote connection signs in with OAuth the first time you use it. You can scope it to one organization with /mcp/<organization> or one project with /mcp/<organization>/<project>. Sentry says scoped tools default to that project and hide the discovery tools you no longer need. It recommends a project scope when you can (mcp.sentry.dev, read November 16, 2026).
| Agent | Command or config | Notes |
|---|---|---|
| Claude Code | claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/<organization>/<project> | Or Sentry's plugin: claude plugin marketplace add getsentry/sentry-mcp, then claude plugin install sentry-mcp@sentry-mcp. The README says the plugin adds a sentry-mcp subagent that Claude hands Sentry questions to. |
| Codex | codex mcp add sentry --url https://mcp.sentry.dev/mcp | Sentry's page has no Codex line; it says any MCP client can use the HTTP endpoint. This is Codex's own command for a remote server, the same form PostHog's docs give for its server. Sentry's pages don't show it working with Sentry's OAuth sign-in, and we did not run it. |
| Cursor | The "Install MCP Server" button on mcp.sentry.dev, or in your MCP settings: {"mcpServers": {"sentry": {"url": "https://mcp.sentry.dev/mcp/<organization>/<project>"}}} | OAuth opens in the browser on first use. |
| VS Code | Command Palette, "MCP: Add Server", HTTP, then the endpoint | From mcp.sentry.dev. |
| Local, any agent | curl https://cli.sentry.dev/install -fsS | bash, sentry auth login, then claude mcp add sentry -- sentry mcp | Runs the server on your machine through the Sentry CLI and its login, so the config holds no token. sentry mcp my-org/my-project limits it to one project. The older npx @sentry/mcp-server package still ships the same server. |
Add ?experimental=1 to the endpoint for what Sentry calls forward-looking tool variants. If the agent shows the server as anything other than connected, how to check if an MCP server is running reads each status line.
Which Sentry MCP tools does the agent get?
Sentry groups its tools into skills, and only two are on by default. These counts come from skillDefinitions.json in the getsentry/sentry-mcp repository, read November 16, 2026. One tool can sit in more than one skill, so the counts overlap.
| Skill | On by default | Tools | What it covers, in Sentry's words |
|---|---|---|---|
| Inspect Issues & Events | Yes | 48 | Read-only access to issues, events, traces, replays, releases, cron, uptime and metric monitors, profiles, documentation and project metadata |
| Seer | Yes | 18 | Sentry's AI debugger that helps you analyze, root cause and fix issues |
| Triage Issues | No | 26 | Resolve, assign and update issues |
| Manage Projects & Teams | No | 25 | Create and change projects, teams, DSNs, monitors and alert rules |
| Documentation | No | 7 | Deprecated; docs tools now sit under Inspect |
The README says the server is built for coding agents with a human in the loop, with tools picked for debugging rather than all of Sentry. With a token on the remote server, ?skills=inspect,triage or ?disable-skills=seer narrows the list. When you run it locally, the AI-powered search tools (search_errors, search_traces, search_logs, search_issues) need your own LLM provider key: OpenAI, Azure OpenAI, Anthropic or OpenRouter.
Can the Sentry MCP server read session replays?
Yes, two tools do, and both depend on Sentry Session Replay already recording your site. From the source files, read November 16, 2026:
search_replays: "Search Sentry session replays by rage clicks, dead clicks, visited pages, errors seen, and users. Returns replay lists, not counts or aggregates."get_replay_details: a replay's summary and a timeline of what the user did: page loads and navigations, clicks including dead and rage clicks, inputs and key presses, console warnings and errors, failed requests and feedback. It is in the Inspect skill and needs theorg:read,project:readandevent:readscopes.
So your agent can pull a list of sessions with rage clicks and read one session's timeline. It returns lists, not a count of how often a fault happens or which one costs you the most signups, so the agent still has to decide what to fix first.
Dead clicks turn up before anyone records anything. When we scanned 493 YC startup homepages on October 28, 2026, 33 (6.7%) had at least one dead click, and 43 of the 49 dead clicks were on real buttons and links (our dead clicks report). Sentry's replay tools find these only in sessions it has already recorded. To check your own page today, the free dead click finder marks each click worked, slow, dead or error.
Who can use it, and what are the limits?
- Sign-in: OAuth for every remote connection. Clients that can send custom headers can pass a Sentry token instead, as
Authorization: Sentry-Bearer <token>; the README says the server does not store, check or refresh that token. - Token scopes: for a User Auth Token the README lists
org:read,project:read,project:write,team:read,team:writeandevent:write. - Write access: Triage and Manage Projects & Teams, the skills that change things, are off by default.
- Self-hosted Sentry: log in with
sentry auth login --urland setSENTRY_HOST. Seer is not part of self-hosted Sentry, so its skill is left out there. - Rate limits: the README says tool calls are subject to the Sentry API’s rate limits. The server retries rate-limited requests after a short wait, and a longer limit returns a Rate Limited message that says when to try again. Agents hit it most when they run many Sentry tool calls in parallel.
What does the Sentry MCP server cost?
Sentry's MCP page and pricing docs list no price for the server. What the agent reads is billed under Sentry's plans. Our pricing ledger read Sentry's free Developer plan (one user, 50 replays) and its Team plan on November 16, 2026; Team is $26 a month billed annually with 50 replays included. Sentry's pricing docs list the replay rates past that:
| Replays a month | Reserved, per replay | Pay-as-you-go, per replay |
|---|---|---|
| Over 50 to 5,000 | $0.0030 | $0.00375 |
| Over 5,000 to 100,000 | $0.00285 | $0.003563 |
At 5,000 replays a month on Team billed annually with reserved replays, that is $40.85, and at 50,000 it is $169.10 (arithmetic ours from those rates, plan price included; errors, spans and everything else Sentry bills left out). Every tier and the monthly plan are on our Sentry pricing page.
Seer, one of the two skills on by default, is a separate charge: Sentry's pricing docs say $40 per active contributor a month until December 8, 2026, then a move to usage-based credits paid from a Seer Wallet with a daily spend limit (read November 16, 2026).
| Plan | Price | Recorded visits a month |
|---|---|---|
| Free | $0, 3 sites | 5,000 |
| Pro | $29 a month | 50,000 |
| Scale | $99 a month | 250,000 |
The two are not like for like: Sentry sells error monitoring with replays beside it, and UserProof sells one thing. Compare the replay line only if replays are the part you need.
Where does Sentry win?
Sentry does far more than UserProof. Its server reaches errors with stack traces, traces and spans, logs, profiles, releases, uptime and cron monitors, and Seer's root-cause analysis, from the same tool your app already reports crashes to. If your problem is code that throws, a slow endpoint or a bad release, this is the server to connect, and no replay tool replaces it. A team already on Sentry gets replays next to the error that caused them with nothing new to install.
What does the Sentry MCP server not tell a founder fixing conversion?
Sentry's tools return lists of replays and one session's timeline. They do not describe a ranked list of where real visitors leave your signup or checkout, tied to the file and line of your code to change, and the replay tools need Session Replay recording set up and paid for first. A page with no error can still lose visitors.
UserProof does that other job and nothing else. It records real visitor sessions with a script tag, and its MCP server runs locally from a plugin. It hands your coding agent one fix at a time with the file and line, ranked by your signup or checkout goal, then checks whether the shipped fix worked. It has no error monitoring, tracing, logs or release tracking. Field values are never recorded. In Claude Code:
claude plugin marketplace add GetUserProof/userproof && claude plugin install userproof@userproof && claude "/userproof:setup"
It also works from Codex and Cursor, as the session replay MCP server page explains. For the two side by side, read UserProof vs Sentry; for other options, Sentry alternatives. How the replay tools with an MCP server compare, each endpoint checked, is in which session replay tools have an MCP server. The same setup for other tools: Datadog, Amplitude and PostHog. More servers worth wiring in are in the best MCP servers for coding agents.
Sources, read November 16, 2026
- Sentry, Sentry MCP Server (https://mcp.sentry.dev/): endpoint, OAuth, scoping, per-client setup, the local CLI server
- Sentry, getsentry/sentry-mcp README (https://github.com/getsentry/sentry-mcp): Claude Code plugin, token header, scopes, LLM keys for local AI search, self-hosted
- Sentry, skillDefinitions.json, search-replays.ts and get-replay-details.ts: skills, tool counts, the replay tools' descriptions and scopes
- Sentry, Pricing & Billing (https://docs.sentry.io/pricing/): replay rates, Seer pricing and the December 8, 2026 change
- PostHog, Codex setup: the
codex mcp add <name> --urlform used in the Codex row - UserProof pricing ledger, Sentry Developer and Team rows read November 16, 2026 (Sentry pricing)
- UserProof: dead clicks on 493 YC homepages (scan of October 28, 2026), plans, set up UserProof in Claude Code
We have no Sentry account, so every step above is quoted from Sentry's pages and source rather than run here.
FAQ
Does Sentry have an MCP server?
Yes. Sentry runs a hosted MCP server at https://mcp.sentry.dev/mcp that signs in with OAuth on first use, and you can scope it to one organization or project. It can also run on your machine through the Sentry CLI with sentry mcp. Sentry says it is built for coding agents with a human in the loop (read November 16, 2026).
How do I add the Sentry MCP server to Claude Code?
Run claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/ followed by your organization and project slugs, then sign in when prompted. Or install Sentry's plugin with claude plugin marketplace add getsentry/sentry-mcp and claude plugin install sentry-mcp@sentry-mcp. This comes from Sentry's pages; we did not run it.
Is the Sentry MCP server free?
Sentry's MCP page and pricing docs list no price for the server itself. What it reads is billed: the Team plan was $26 a month billed annually with 50 replays on November 16, 2026, with replays past that from $0.0030 each, and Seer costs $40 per active contributor a month until December 8, 2026, then moves to usage-based credits.
Can the Sentry MCP server show my coding agent session replays?
Yes, if Sentry Session Replay is already recording your site. search_replays lists replays by rage clicks, dead clicks, pages visited, errors seen and users, and returns lists rather than counts. get_replay_details returns one replay's summary and a timeline of page loads, clicks, inputs, console errors, failed requests and feedback.
Can the Sentry MCP server change things in my Sentry account?
Only if you turn it on. The skills that change things, Triage Issues (resolve, assign, update) and Manage Projects & Teams (projects, teams, DSNs, monitors, alert rules), are off by default. Inspect, which is read-only, and Seer are on by default, according to the skill definitions in Sentry's source (read November 16, 2026).
What can't the Sentry MCP server tell me about conversion?
Its tools return lists of errors, traces and replays and one session's timeline. They do not describe a ranked list of where visitors drop off tied to the file and line of your code to change. UserProof does that one job: it records real visitor sessions and hands your coding agent one fix at a time with the file and line.