Developer Tools

Sentry MCP server: scoped access for every coding agent

Sentry MCP is the official Model Context Protocol server from Sentry, hosted at mcp.sentry.dev, that lets Claude Code, Cursor, Codex and other AI agents search issues, read events and traces, and run Seer analysis over OAuth. AIPx connects that server once and gives each agent its own URL, with reads on Allow and issue updates behind Confirm.

Updated 14 October 2026 9 highlighted tools 6 read · 3 write

What your agent can do with Sentry

Nine direct tools, six of them reads. The other 48 of the server's 57 operations sit behind search_sentry_tools and execute_sentry_tool.

Tool	What it does	Access	Mode
find_organizations	Your organizations	read	Allow
find_projects	Projects in an organization	read	Allow
search_issues	Search issues in plain language or Sentry syntax	read	Allow
search_events	Search events and replays, with counts	read	Allow
get_sentry_resource	Fetch an issue, event or trace by URL or ID	read	Allow
search_sentry_tools	Search the remaining operations	read	Allow
analyze_issue_with_seer	Seer root cause; starts a 2–5 minute run if none exists	write	Confirm
update_issue	Resolve, reopen, assign or ignore	write	Confirm
execute_sentry_tool	Run a catalog operation, read or write	write	Hide

Sentry exposes a compact direct surface plus a searchable catalog of long-tail operations. The visible set varies with Sentry skills and account capabilities.

Capability snapshot: 14 October 2026 · provider source

Look twice at the last row: one mode covers everything from reading a dashboard to delete_uptime_monitor. Source: getsentry/sentry-mcp, version 0.39.0.

Why give Claude Code and Cursor Sentry MCP access

A pasted stack trace drops the release, environment, tags and breadcrumbs. With read tools the agent inspects the issue in place and ties the failing frame to your code. Resolving it is a separate decision. The same URL also carries Supabase MCP and all AIPx integrations.

Connect Sentry MCP to Claude, Claude Code, Cursor or Codex

mcp.sentry.dev is a remote MCP server; per-client setup is in the official Sentry MCP documentation. With AIPx as your MCP gateway, you log in to Sentry once.

Step 1: connect Sentry in AIPx

In dash.aipx.run, open Integrations → Sentry and finish the MCP OAuth flow.
On Sentry’s consent screen, Inspect and Seer are ticked by default. Tick Triage Issues only if an agent needs update_issue.
Create an MCP agent, turn on Sentry, set modes, save.
Copy the endpoint: https://dash.aipx.run/mcp/<agent>.

Step 2 is the usual trap. Skip Triage Issues and update_issue never shows up, whatever mode AIPx has.

Sentry MCP OAuth screen with Inspect and Seer skills enabled

Sentry MCP in Claude Code

claude mcp add --transport http --scope user aip https://dash.aipx.run/mcp/<agent>

Then /mcp, pick aip, authorize. More in Claude Code MCP setup.

Add Sentry MCP to Claude Code through an AIPx agent URL

Sentry MCP in Cursor

In ~/.cursor/mcp.json:

{
  "mcpServers": {
    "aip": { "url": "https://dash.aipx.run/mcp/<agent>" }
  }
}

See Cursor MCP.

Sentry MCP in Codex

codex mcp add aip --url https://dash.aipx.run/mcp/<agent>
codex mcp login aip

See Codex MCP.

Connect Claude to Sentry

Customize → Connectors → Add custom connector, paste the agent URL. Claude Free allows one custom connector; one AIPx URL fits all your accounts into it. See Claude MCP connectors and Claude Desktop MCP.

Direct Sentry MCP vs through AIPx

	Direct	Through AIPx
Setup	OAuth in each client	One Sentry OAuth
Permissions	Skills only	Skills plus per-tool, per-agent modes
Other services	No	Notion MCP, Cloudflare MCP and more
Audit	Per client	One log
Self-hosted Sentry	stdio server	Not supported

Honestly, if Sentry only runs in Claude Code, the direct claude mcp add is simpler.

Scope Sentry MCP per agent

Sentry skills decide what exists; AIPx modes decide what each agent sees, never more than your Sentry user has. See per-tool MCP permissions.

Tool	Claude Code	Cursor	Triage bot
Reads	Allow	Allow	Allow
`analyze_issue_with_seer`	Confirm	Hide	Confirm
`update_issue`	Confirm	Hide	Confirm
`execute_sentry_tool`	Hide	Hide	Confirm

Seer stays on Confirm because a loop over new errors can start a dozen runs in an evening, at $40 per active contributor. On the free Developer plan Seer doesn’t work, so hide it.

One limitation: AIPx uses the unscoped https://mcp.sentry.dev/mcp, so no /{org}/{project} pinning. Split by agent. The MCP server catalog shows what else agents can carry.

Sentry MCP server tool permissions in AIPx: Allow, Confirm, Hide

Two Sentry accounts in one connector

A personal Sentry and a client’s organization fit in one connector, with modes per tool × account × agent. Free: 3 accounts, 1 agent. Pro: $3/month (AIPx pricing).

Close the loop without auto-closing the issue

Let the agent read Sentry, test a fix, open a Linear MCP or Jira MCP ticket, check impact in PostHog MCP. The final update should follow evidence, not confidence.

But Confirm is a hint, not a server-side pause. If the client auto-runs tools, the call goes through. Keep client confirmations on and use Hide for anything risky (MCP security).

Sentry MCP calls in the AIPx audit log

The Sentry official MCP documentation describes the direct vendor connection and its authorization requirements.

AIPx permission matrix for two Gmail accounts, with per-tool Allow, Confirm and Hide modes — Two Gmail accounts on one agent, each with its own Allow, Confirm and Hide rules. Allow and Confirm are client execution hints; Hide removes the tool from this agent.

Route Sentry context to the agents that can use it

Keep searches available while triage updates wait for approval and long-tail operations stay hidden.

Connect Sentry free

What to ask your agent

Ask for evidence before any state change:

> Find new production issues in checkout during the last 24 hours and rank them by affected users.

> Pull the latest events for this issue, identify the first application frame, and trace it to the likely code path.

> Ask Seer to analyze the issue, then separate its hypotheses from facts visible in the event data.

> Propose the correct owner and resolution state for this issue, but wait before updating Sentry.

FAQ

What is Sentry MCP?

Sentry's official Model Context Protocol server at https://mcp.sentry.dev/mcp. After OAuth, AI agents can search issues and events, fetch traces and run Seer.