Sourcegraph

@sourcegraph list all repositories in our org that use TypeScript, sorted by most recent commit

Based on the Sourcegraph instance query, 47 repositories use TypeScript. The five most recently updated are: `web-platform` (last commit 2 hours ago), `api-gateway` (last commit 5 hours ago), `design-system` (last commit 1 day ago), `analytics-service` (last commit 2 days ago), and `mobile-app` (last commit 3 days ago). Each repository shows TypeScript as either primary or secondary language. This list reflects repositories visible to your authenticated user account.

This example combines the list-repositories and list-languages tools to filter by stack. Useful for auditing codebases or finding migration candidates. Requires read access to all repositories you want to scan; private repos outside your scope won't appear. Large instances may hit pagination limits—results reflect the first page of matches.

@sourcegraph compare commits abc123 and def456 in the api-gateway repo and summarize what changed in the authentication layer

Comparing commits abc123 and def456 in `api-gateway` shows 8 files changed. In the authentication layer: `src/auth/jwt.ts` added token expiry validation (lines 45-62), `src/auth/middleware.ts` refactored error handling to return 401 instead of 500 for invalid tokens, and `tests/auth.test.ts` added three new test cases for edge conditions. The diff indicates a security hardening effort focused on token lifecycle management. No breaking changes to the public API surface.

This example uses the compare-commits tool to inspect code evolution, then layers AI reasoning to summarize intent. Requires knowing exact commit SHAs beforehand—use get-commit-details first if you only have branch names. Diffs for large commits may be truncated; the MCP returns raw diff text that the AI interprets.

@sourcegraph fetch the contents of src/config/database.ts from the api-gateway repo and check if connection pooling is enabled

Retrieved `src/config/database.ts` from the default branch of `api-gateway`. The file exports a `createPool` function that initializes a connection pool with `max: 20` and `idleTimeoutMillis: 30000`. Connection pooling is enabled. The configuration uses the `pg` library and reads pool size from the `DB_POOL_SIZE` environment variable, falling back to 20. No SSL enforcement is visible in this file—check `src/config/ssl.ts` if that's a concern.

This example demonstrates file-content retrieval paired with AI analysis—useful for quick config audits without cloning repos. The MCP fetches from the default branch only; specify a different branch in the prompt if needed. Large files may be truncated. Requires read access to the repository.

When Sourcegraph beats grep for multi-repo teams

A 12-person engineering team maintaining six microservices needs to audit which repos still import a deprecated auth library before the Q2 migration deadline. The Sourcegraph MCP wins here because List Repositories plus Get File Contents lets you scan package.json or requirements.txt files across all repos in under two minutes without cloning anything locally. Compare this to GitHub's MCP, which requires separate API calls per repo and hits rate limits around repo 40. The threshold: if you're only checking 2-3 repos, just open them in your editor. But once you're auditing 5+ repos or searching for patterns across a codebase you don't own, Sourcegraph's indexed search makes this a 90-second Switchy prompt instead of a 20-minute manual slog.

Sourcegraph as the new-hire documentation layer

A startup's third backend engineer joins and needs to understand how the payment flow works across the monorepo's 18 modules. The Sourcegraph MCP is the right call when your team doesn't maintain architecture docs and the code is the source of truth. Use List Repository Files to map the directory structure, then Get File Contents to pull specific handlers or config files into the conversation. The AI can trace the flow without the new hire needing repo access or IDE setup on day one. This breaks down if your codebase is under 5k lines—just read it directly—or if it's over 500k lines and poorly organized, where even indexed search returns too many false positives. Sweet spot: 20k-200k line repos where the structure is discoverable but not trivial.

When Compare Commits clarifies the production break

A 6-person SRE team is debugging why the API started timing out after last night's deploy. The Sourcegraph MCP's Compare Commits tool is faster than GitHub's web UI when you need to see diffs for 40+ changed files in a single conversation context. Pull the commit range, feed the diffs to the AI, and ask 'which of these changes could cause a 3-second delay in the /orders endpoint'. This works because Sourcegraph returns structured diffs the AI can parse, not just links. The trade-off: if the incident spans multiple repos or you need to correlate with logs, you'll still need your observability stack. Use this when the smoking gun is in the code and you need to triage it in Slack during the incident call, not after.