https://expensify.slack.com/archives/C05LX9D6E07/p1774429406933519
Strategy: Engineering velocity is one of the highest-leverage inputs to developer productivity. Reducing the friction of common diagnostic workflows — especially those that interrupt flow state — directly improves the throughput of the team without increasing headcount. Tooling that eliminates context-switching between an AI assistant and external dashboards compresses the feedback loop on every investigation.
Background: The Expensify engineering team uses Sentry as the primary platform for performance profiling, and production incident tracking. Claude Code is the team's AI-powered development assistant, already extended with external tools via MCP (Model Context Protocol) — for example, the Playwright MCP is configured in .mcp.json and surfaces browser automation tools directly inside the assistant. When engineers investigate bugs or production incidents, they regularly need to search for error patterns, review stack traces, and correlate Sentry event data with source code. All of this data lives in Sentry.
Problem: When engineers are debugging a production issue, if they attempt to correlate Sentry error data with their local codebase inside Claude Code, then they must manually context-switch to the Sentry web UI, copy relevant data, and return to their assistant — fragmenting the investigation and increasing time-to-resolution.
Solution: Add the Sentry MCP server to the project's .mcp.json using Sentry's hosted OAuth remote endpoint. Claude Code handles authentication automatically via browser OAuth on first use — no personal token creation or environment variable management required. This surfaces Sentry tools directly in Claude Code: search issues by query, retrieve stack traces and event details, inspect distributed traces, and run AI-powered root cause analysis. Add corresponding permissions to .claude/settings.json so tools run without per-call confirmation prompts, matching the existing Playwright MCP pattern. Document the one-time OAuth flow in CLAUDE.md.
With this in place, an engineer can ask Claude Code to "find recent crashes in the expense creation flow" or "analyze this Sentry issue" without leaving their development context. Sentry data feeds directly into the investigation, enabling Claude Code to correlate error patterns with the relevant source code in a single session.
Issue Owner
Current Issue Owner: @sosek108
https://expensify.slack.com/archives/C05LX9D6E07/p1774429406933519
Strategy: Engineering velocity is one of the highest-leverage inputs to developer productivity. Reducing the friction of common diagnostic workflows — especially those that interrupt flow state — directly improves the throughput of the team without increasing headcount. Tooling that eliminates context-switching between an AI assistant and external dashboards compresses the feedback loop on every investigation.
Background: The Expensify engineering team uses Sentry as the primary platform for performance profiling, and production incident tracking. Claude Code is the team's AI-powered development assistant, already extended with external tools via MCP (Model Context Protocol) — for example, the Playwright MCP is configured in .mcp.json and surfaces browser automation tools directly inside the assistant. When engineers investigate bugs or production incidents, they regularly need to search for error patterns, review stack traces, and correlate Sentry event data with source code. All of this data lives in Sentry.
Problem: When engineers are debugging a production issue, if they attempt to correlate Sentry error data with their local codebase inside Claude Code, then they must manually context-switch to the Sentry web UI, copy relevant data, and return to their assistant — fragmenting the investigation and increasing time-to-resolution.
Solution: Add the Sentry MCP server to the project's .mcp.json using Sentry's hosted OAuth remote endpoint. Claude Code handles authentication automatically via browser OAuth on first use — no personal token creation or environment variable management required. This surfaces Sentry tools directly in Claude Code: search issues by query, retrieve stack traces and event details, inspect distributed traces, and run AI-powered root cause analysis. Add corresponding permissions to .claude/settings.json so tools run without per-call confirmation prompts, matching the existing Playwright MCP pattern. Document the one-time OAuth flow in CLAUDE.md.
With this in place, an engineer can ask Claude Code to "find recent crashes in the expense creation flow" or "analyze this Sentry issue" without leaving their development context. Sentry data feeds directly into the investigation, enabling Claude Code to correlate error patterns with the relevant source code in a single session.
Issue Owner
Current Issue Owner: @sosek108