What is Sentry?

• Error & performance monitoring — errors, traces/spans, logs, session replay, releases
• You ship code → Sentry tells you what broke, where, and why (down to the stack frame)
• Traditionally lived in a web dashboard

Today's talk: getting that data out of the browser — into your terminal, your code, and your AI agents.

Meet the Sentry CLI

A first-class CLI — "built by robots and humans, for robots and humans."

• Native binaries (curl / Homebrew / npm), Node ≥ 22
• Dedicated commands for what you actually do:

sentry issue list
sentry issue view ACME-123
sentry event view   sentry trace view   sentry log list
sentry api  /  sentry schema   # anything without a dedicated command

• Two output modes baked in: pretty for humans, --json for machines

sentry init: zero to instrumented

sentry init                 # auto-detect everything
sentry init acme/my-app     # explicit org/project
sentry init ./my-project    # point at a directory

• Runs the setup wizard: detects your framework, installs the SDK, wires up config
• Pick features, or pass them: --features errors,tracing,replay
• Agent/CI friendly — non-interactive:

sentry init --yes --features errors,tracing
sentry init --dry-run        # preview before touching files

The CLI is a library (npm)

npm install sentry ships both the CLI and a typed SDK — no subprocess.

import createSentrySDK from "sentry";

const sdk = createSentrySDK({ token: "sntrys_..." });

// Typed methods for every command
const issues = await sdk.issue.list({ orgProject: "acme/frontend", limit: 5 });
const issue  = await sdk.issue.view({ issue: "ACME-123" });

// Escape hatch for anything
const version = await sdk.run("--version");

• In-process (no spawn) · parsed objects (zero-copy) · typed SentryError
• Never touches your process.env; streaming commands return async iterables

Same code path as the CLI — just callable. Great for build tools, CI, and agents.

Seer — Sentry's built-in agent

Sentry already has an AI agent that knows your stack traces, traces, and code. The CLI puts it one command away:

sentry issue explain @latest      # root-cause analysis
sentry issue plan ACME-123        # step-by-step fix plan

• explain → root causes, reproduction steps, relevant code locations
• plan → concrete implementation steps (uses your GitHub integration + code mappings)
• Works on real issues — @latest, @most_frequent, or a short ID like cli-G

explain/plan start from a known issue.

Seer Agent + agentic dashboards

Seer Agent (open beta) — when there's no issue to start from, just ask:

• Hit Cmd + / / "Ask Seer" in Sentry, or @mention it in Slack
• Traverses your trace-connected graph (trace → spans → logs → deploy → commits → source) — no "which service is upstream?" guesswork

Agentic dashboards — built from the same engine:

• Prompt "Generate Dashboard" in-product, with full revision history to revert mistakes
• Or script it: sentry dashboard create + --json so agents (Claude Code, Copilot, Cursor) provision dashboards from your terminal

Explore rewards operators who already have the map. Seer builds the map.

Agentic Sentry — two paths

1. Sentry CLI as an agent skill

• On install, sentry cli setup drops skills into ~/.claude / ~/.agents (Claude Code, Cursor…)
• Agent translates intent → real commands:
  • "Show me the latest issues" → sentry issue list
  • "Stack trace for ISSUE-123?" → sentry issue view ISSUE-123
• Uses your existing sentry auth login — includes Seer AI (root-cause + fix plans)

Agentic Sentry — two paths

2. Sentry MCP server — mcp.sentry.dev · getsentry/sentry-mcp

• Remote MCP server: middleware to the Sentry API, built for coding assistants
• Connect any MCP client (Cursor, Claude Code…) — OAuth, no token wrangling
• Or run local stdio: npx @sentry/mcp-server@latest --access-token=…
• AI-powered tools: search_events, search_issues, Seer

CLI skill = local, project-aware, scriptable. MCP = remote, account-wide tool surface. They complement each other.