[litellm-agent] Staging → litellm_internal_staging (5/7/2026)

[oss-pr-review-agent-shin]

Automated staging PR created by litellm-agent.

This branch collects PRs approved by the agent on 5/7/2026.

[oss-pr-review-agent-shin]

@greptile please review

[codecov]

Codecov Report

❌ Patch coverage is 87.25490% with 13 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ellm/proxy/_experimental/mcp_server/oauth_utils.py	87.23%	6 Missing ⚠️
...itellm/proxy/health_endpoints/_health_endpoints.py	68.42%	6 Missing ⚠️
litellm/proxy/auth/handle_jwt.py	88.88%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[oss-pr-review-agent-shin]

@greptile please review

[oss-pr-review-agent-shin]

@greptile please review

[greptile-apps]

Greptile Summary

This automated staging PR bundles five independently-reviewed fixes: OAuth redirect-URI validation extended to same-origin (UI callback flow), JWT team-ID claim union for Okta/Auth0 single-team users, guardrail provenance tracking for cross-pod stale-entry reconciliation, deployment disambiguation by model_info.id in the health test-connection endpoint, and a server-side-filtering refactor for the Virtual Keys table.

• MCP OAuth (oauth_utils.py, discoverable_endpoints.py): validate_trusted_redirect_uri replaces the loopback-only guard, adding same-origin acceptance for the proxy's own UI callback while preserving loopback for native MCP clients. get_request_base_url is de-duplicated into oauth_utils; X-Forwarded-* headers are only trusted from configured proxy ranges.
• JWT + Guardrails (handle_jwt.py, guardrail_registry.py, guardrail_endpoints.py, proxy_server.py): get_all_jwt_team_ids unions both plural and singular IdP claim fields; the guardrail registry gains a _sources provenance map and reconcile_db_guardrails to purge entries deleted on another pod, with list/get-by-id endpoints updated to surface correct 404s for stale in-memory rows.
• Frontend (VirtualKeysTable.tsx, filter_logic.tsx, model_info_view.tsx): filtering delegated from the client-side hook to the useKeys query params, with 300 ms debounce and page-reset on filter change; model_info.id now forwarded to the test-connection endpoint to disambiguate wildcard multi-deployment models.

Confidence Score: 5/5

Safe to merge — all changed paths are well-scoped, thoroughly mocked, and guarded by new regression tests.

Each change is narrow in scope with direct test coverage. The OAuth redirect validation tightens the trust boundary; the JWT union is additive only; the guardrail provenance system correctly segregates config from DB entries and never removes config-loaded guardrails during reconciliation; the health endpoint disambiguates only when an id is explicitly supplied. No backwards-incompatible interface changes, no direct DB queries in new hot paths, and all new tests are properly mocked.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/proxy/_experimental/mcp_server/oauth_utils.py	Adds get_request_base_url (moved from discoverable_endpoints) and validate_trusted_redirect_uri that accepts same-origin or loopback URIs; X-Forwarded-* headers only honoured from trusted proxy ranges
litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py	Replaces loopback-only redirect_uri validation with validate_trusted_redirect_uri; callback now receives Request param for same-origin origin check; get_request_base_url de-duplicated into oauth_utils
litellm/proxy/auth/handle_jwt.py	Adds get_all_jwt_team_ids that unions plural (team_ids_jwt_field) and singular (team_id_jwt_field) claims, deduplicated, without consulting team_id_default
litellm/proxy/guardrails/guardrail_registry.py	Adds _sources provenance dict (db/config), get_source(), and reconcile_db_guardrails() to drop stale cross-pod DB entries; source param threads through initialize/update/reinitialize paths
litellm/proxy/guardrails/guardrail_endpoints.py	list_guardrails_v2 skips DB-backed in-memory entries absent from DB query; get_guardrail_info returns 404 for stale DB-backed entries; create/approve calls pass source=db
litellm/proxy/health_endpoints/_health_endpoints.py	test_model_connection now disambiguates by model_info.id (get_deployment) first, falling back to model_name lookup; fixes silent use of deployments[0] for wildcard multi-deployment models
litellm/proxy/proxy_server.py	_init_guardrails_in_db now collects db_guardrail_ids set and calls reconcile_db_guardrails after syncing, purging stale entries deleted on another pod
litellm/proxy/management_endpoints/ui_sso.py	SSO callback now calls get_all_jwt_team_ids instead of get_team_ids_from_jwt in two places, picking up both singular and plural claim fields
ui/litellm-dashboard/src/components/VirtualKeysPage/VirtualKeysTable.tsx	Moves filtering from client-side (filter_logic hook) to server-side (useKeys params); debounces filter inputs before API calls; pagination resets on filter change
ui/litellm-dashboard/src/components/key_team_helpers/filter_logic.tsx	Simplified to pure filter-state management; removes debouncedSearch API call, filteredKeys, filteredTotalCount, Sort By/Sort Order fields; adds Key Hash field
ui/litellm-dashboard/src/components/model_info_view.tsx	Passes model_info.id in the test-connection request payload so the backend can disambiguate between deployments sharing a wildcard model_name
litellm/proxy/guardrails/init_guardrails.py	Passes source=config when initializing guardrails from YAML config file

_{Reviews (2): Last reviewed commit: "fix(ui): virtual keys filter silently re..." | Re-trigger Greptile}

[oss-pr-review-agent-shin]

@greptile please review

[oss-pr-review-agent-shin]

@greptile please review