Add a keyless Parallel Search MCP example

[georgeatparallel]

Summary

Add a runnable example for free web search and page fetching with Parallel Search MCP, using HarnessX's existing MCPToolRegistry. Search runs in Fast mode without an account or API key. Install from the repository root and run examples/parallel_search/search.py with an objective, --query, and optional --fetch-url arguments.

Type of Change

• New feature
• Docs

Checklist

• pytest tests/unit tests/integration passes locally
• No hardcoded API keys, internal addresses, or tokens
• New files confirmed to contain no sensitive information
• pyproject.toml updated if new dependencies were added
• Frontend changes pass npm run build (not applicable)

Notes for Reviewer

HTTP headers are optional; existing provider and saved configuration defaults remain unchanged. The MCP dependency stays on 1.x because the current transport uses its streamablehttp_client API. The example calls tools directly and is subject to the endpoint's free-tier rate limits.

I work at Parallel.

[zhao9797]

Reviewed the final head 3b529d1. The optional HTTP headers preserve existing call behavior, and the example handles connection cleanup on success and failure.

Verified real anonymous web_search and web_fetch calls with MCP 1.30.0: both returned content and the example exited successfully. The lockfile now matches mcp>=1.30,<2 and includes the previously declared html2text dependency; uv lock --check passes.

All five required checks pass on this final commit: Python 3.11 and 3.12 unit/integration tests, Ruff, pre-commit, and frontend build. Local macOS testing also found one existing /etc/passwd versus /private/etc/passwd assertion failure, reproduced on main and unrelated to this change.

Approved for merge.