Python: feat: Add Serpex multi-engine web search connector

[Kartey R (divyeshradadiya)]

---

title: "Add Serpex Search Connector"
labels: "area/connectors, enhancement"
assignees: ""

Summary

This PR adds a new Serpex search connector that implements the TextSearch interface and provides async search support across Serpex engines (auto, google, bing, duckduckgo, brave, yahoo, yandex). The connector integrates with the Semantic Kernel connector patterns and includes unit tests that mock the HTTP client.

Motivation

• Provide a simple, tested adapter for the Serpex search API so developers can use Serpex as a search provider inside Semantic Kernel.
• Support multiple engines via the engine parameter with sensible defaults.
• Follow existing connector conventions (pydantic models, async search, get_text_search_results, get_search_results).

Design

• New connector: semantic_kernel.connectors.serpex.SerpexSearch

  • Config model uses KernelBaseModel/KernelBaseSettings patterns (pydantic) for typed configuration.
  • Uses httpx.AsyncClient for HTTP calls (same pattern as other connectors).
  • Exposes async search which returns structured search results and helper methods for text-only or raw results.

• Error handling:

  • Gracefully raises connector-specific exceptions on HTTP error responses.
  • Validates API key and inputs; surfaces Serpex error messages when available.

Files Changed (high level)

• python/semantic_kernel/connectors/serpex.py — New connector implementation.
• python/tests/unit/connectors/search/test_serpex_search.py — Unit tests for success and error flows.

Usage

Example (async):

from semantic_kernel.connectors.serpex import SerpexSearch

# instantiate with an API key
connector = SerpexSearch(api_key="sk_...")

# perform a search (async)
results = await connector.search(query="semantic kernel connectors", engine="google", num=10)

# get text-only results
text_results = await connector.get_text_search_results(query="semantic kernel connectors")

Notes:

• engine defaults to auto if not provided.
• api_key should be provided as a secret in production (use env var / secrets manager).

Tests

• Unit tests included:
  • test_serpex_search.py covers successful search response parsing and error response handling.
  • Tests mock httpx.AsyncClient to avoid network calls and follow existing test patterns in the repo.

Compatibility

• Connector uses pydantic models present in the repository and httpx.AsyncClient.
• No breaking changes to public APIs.
• Please run the repository's formatting/linting (black/ruff) and type checks (mypy) per CONTRIBUTING.md.

Checklist for reviewers

• Code style and formatting (black/ruff).
• Types / type annotations (mypy).
• Tests: unit tests present and follow mock patterns.
• No secrets committed.
• Documentation: brief usage example included in PR description.
• Confirm connector is discoverable (exports if required by the project layout).

Security & Privacy

• The connector expects the Serpex API key. Do NOT commit API keys. Use environment variables or secret management.
• The connector forwards search queries to Serpex; ensure compliance with your organization's data policies.

[Kartey R (divyeshradadiya)]

@microsoft-github-policy-service agree company="serpex.dev"

[Evan Mattson (moonbox3)]

Hi Kartey R (@divyeshradadiya), thank you for your contribution. Current feature focus is on SK's successor SDK - Microsoft Agent Framework. Please have a look at the repo if you haven't already. https://github.com/microsoft/agent-framework