Sentry
by Sentry
Pull production error events, stack traces and crash attachments into a coding agent. The sharp edge is a read tool, not a write one: get_event_attachment embeds screenshots and minidumps directly in the model's context. Sentry's server-side scrubber is on by default for events, but attachments are opt-in only.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Sentry via MCP
https://mcp.sentry.dev/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Use in Agentman
Connect Sentry once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioSentry Tools & Capabilities (20)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- The read side is the sharp side.
get_event_attachmentandget_issue_detailsare annotated read-only and can still pull real end users' data into a context window. Review this connector on what its reads expose, not only on its write label. - Attachment scrubbing is opt-in and easy to get wrong. Sentry documents limited support for non-minidump attachments: only rules naming a specific filename scrub, and a wildcard rule does not. Audit your org's rules before granting attachment reads.
- Event scrubbing is a keyword denylist. It is on by default and covers a published field list, but it matches fixed strings. Sensitive values under unconventional key names need Additional Sensitive Fields or an Advanced Data Scrubbing rule.
- Geo data outlives IP suppression. Sentry states geographic information is extracted from the user's IP even when IP storage is disabled, and needs its own advanced rule to remove.
update_issuecan mute a live alert. Resolving or ignoring an issue stops alerting on it. Reversible, and logged to the activity feed when a reason is supplied — but a suppressed page is not undone by a later fix.- Four directory tool names are stale.
find_issues,find_tags,find_errorsandfind_transactionshave no counterpart in the current open-source catalogue. - The directory tool count is under half the real surface. 20 listed against 52 defined.
- Annotations are from source, not the wire. We could not run an authenticated
tools/list, so the deployed server could differ frommain. - Prompts are unconfirmed. Anthropic's snapshot lists none, and the gated endpoint blocked our own
prompts/list. - The directory's documentation URL is dead.
docs.sentry.io/product/sentry-mcp/returns a Page Not Found. The live documentation is atmcp.sentry.devand the GitHub README. - The hosted endpoint does not serve self-hosted Sentry. Run the stdio server with a host flag instead, and expect Seer to be unavailable.
- Natural-language search needs an LLM provider when self-hosted. Sentry states those tools are unavailable without one; other tools still work.
plainPKCE is advertised. The MCP authorization-server document offersplainalongsideS256. PreferS256.
Frequently asked questions
Whatever your events captured — request headers, cookies, local variables in stack frames, user email and IP, breadcrumbs, and any file attached to a crash. Sentry's server-side scrubber is on by default and redacts those event fields, but attachments are different: Sentry documents that only attachments named by an explicit non-wildcard rule get scrubbed at all.
No. `update_issue` changes status and assignment only — resolve, reopen, assign, or ignore. Sentry's open-source server defines an `event:admin` scope for deleting issues, but the MCP server never requests it and no catalogue tool issues an issue deletion. Sentry's docs note single events cannot be deleted from the UI either, only whole issues.
Yes, and this is the real write risk. `update_issue` can set status to resolved or ignored, which stops Sentry alerting on a live error. Ignore modes include `forDuration` and `untilOccurrenceCount`. Sentry annotates the tool `destructiveHint: true` and non-idempotent, and puts it behind a Triage grant that is off by default.
Yes, and unusually we can report real values. Sentry's server is open source, and its tool type makes `readOnlyHint`, `destructiveHint` and `openWorldHint` mandatory, with a test asserting every tool declares them. Across 52 catalogue tools, 38 are annotated read-only and 6 destructive. These come from source, not a live handshake.
Not the hosted endpoint — run your own copy instead. Sentry's README documents a stdio transport taking a `--host` flag pointing at your instance, plus `--insecure-http` for internal deployments without TLS. Sentry notes some features such as Seer may be unavailable self-hosted, and provides a flag to disable those skills so unsupported tools are not exposed.
Four: `org:read`, `project:write`, `team:write` and `event:write`. We read these live from the protected-resource descriptor on 2026-08-19. Sentry's own API advertises 26 scopes at its issuer, so the MCP server takes a narrow slice — notably omitting the admin scopes that delete projects, teams and issues.
Scope the URL and decline the write skills. Sentry documents path-scoped endpoints that pin the connection to one organization or one project, and recommends project scoping where possible. Separately, its skill grants for Triage and Manage Projects & Teams are both off by default in source, so a default connection is read plus Seer.
Sources
- Sentry MCP server documentation — https://mcp.sentry.dev/llms.txt and https://mcp.sentry.dev/ (both fetched 2026-08-19). The directory's
documentationURLdocs.sentry.io/product/sentry-mcp/returns a Page Not Found · retrieved 2026-08-19 - sentry-mcp source repository — https://github.com/getsentry/sentry-mcp (fetched 2026-08-19;
mainbranch, last pushed 2026-08-19). Tool catalogue, annotations, skill defaults and permission model all read from source · retrieved 2026-08-19 - Tool catalogue directory — https://github.com/getsentry/sentry-mcp/tree/main/packages/mcp-core/src/tools/catalog (fetched 2026-08-19; 52 tool definitions) · retrieved 2026-08-19
- Tool type and annotation contract — https://github.com/getsentry/sentry-mcp/blob/main/packages/mcp-core/src/tools/types.ts (fetched 2026-08-19) · retrieved 2026-08-19
- Annotation enforcement test — https://github.com/getsentry/sentry-mcp/blob/main/packages/mcp-core/src/tools/tools.test.ts (fetched 2026-08-19) · retrieved 2026-08-19
- Skill definitions and defaults — https://github.com/getsentry/sentry-mcp/blob/main/packages/mcp-core/src/skills.ts (fetched 2026-08-19) · retrieved 2026-08-19
- Scope hierarchy — https://github.com/getsentry/sentry-mcp/blob/main/packages/mcp-core/src/permissions.ts (fetched 2026-08-19) · retrieved 2026-08-19
update_issuedefinition — https://github.com/getsentry/sentry-mcp/blob/main/packages/mcp-core/src/tools/catalog/update-issue.ts (fetched 2026-08-19) · retrieved 2026-08-19- Sentry Server-Side Data Scrubbing — https://docs.sentry.io/security-legal-pii/scrubbing/server-side-scrubbing/ (fetched 2026-08-19 as Markdown via the
.mdsuffix) · retrieved 2026-08-19 - Sentry Event PII Fields — https://docs.sentry.io/security-legal-pii/scrubbing/server-side-scrubbing/event-pii-fields/ (fetched 2026-08-19) · retrieved 2026-08-19
- Sentry Attachment Scrubbing — https://docs.sentry.io/security-legal-pii/scrubbing/attachment-scrubbing/ (fetched 2026-08-19) · retrieved 2026-08-19
- Sentry JavaScript configuration options — https://docs.sentry.io/platforms/javascript/configuration/options/ (fetched 2026-08-19;
sendDefaultPiidefaults tofalseand is deprecated in favour ofdataCollection) · retrieved 2026-08-19 - Sentry docs machine index — https://docs.sentry.io/llms.txt (fetched 2026-08-19) · retrieved 2026-08-19
docs.sentry.io/robots.txtallows all paths and carriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— synthesis is expressly permitted on all three axes (fetched 2026-08-19) · retrieved 2026-08-19- Live RFC 9728 protected-resource descriptor — https://mcp.sentry.dev/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19; the root form 404s and the path-append form 401s) · retrieved 2026-08-19
- Live RFC 8414 authorization-server metadata — https://mcp.sentry.dev/.well-known/oauth-authorization-server and https://sentry.io/.well-known/oauth-authorization-server (both fetched 2026-08-19; the issuer advertises 26 scopes against the MCP server's 4) · retrieved 2026-08-19
- Live anonymous probe — 2026-08-19:
initializePOST to the endpoint returns HTTP 401 withwww-authenticate: Bearer realm="OAuth", resource_metadata="https://mcp.sentry.dev/.well-known/oauth-protected-resource/mcp"· retrieved 2026-08-19 - Anthropic connector directory — https://claude.ai/directory/46d6322a-5f75-4822-b739-f49261805e9c (snapshot 2026-08-16) · retrieved 2026-08-16
Server Info
- Category
- Developer Tools
- Developer
- Sentry
- Tools
- 20
- Domain
- mcp.sentry.dev
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.
Similar MCP Servers for Developer Tools
Related Resources
Ready to connect Sentry?
Connect Sentry once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.