feat(tool): add opt-in Parallel Search MCP bridge

[georgeatparallel]

Describe this PR

Adds an opt-in Parallel Search MCP tool through MiroFlow's existing stdio MCP path. Parallel Search MCP is free and keyless; the agent's language model retains its normal credential requirements.

What changed?

• Added tool-searching-parallel, a FastMCP bridge to https://search.parallel.ai/mcp, with a MiroFlow User-Agent and optional proxy/CA settings.
• Added agent_quickstart_parallel_search and setup documentation in the tool navigation.
• Added local tests for configuration, discovery, execution, headers, exclusions, errors and a final answer through the orchestrator.

Run uv run main.py trace --config_file_name=agent_quickstart_parallel_search --task="Search for the Python asyncio documentation and explain what asyncio.run does, with a source URL." after configuring the existing OpenRouter key.

Why?

This provides another search and page-fetching option using the existing FastMCP dependency. Existing agents, Serper/Jina settings and tool result handling are preserved.

Related issues

None.

Checklist for PR

• Title follows the Angular commit message format.
• Ran uv tool run ruff@0.8.0 check --fix . and uv tool run ruff@0.8.0 format ..
• Ran uv run python -m unittest discover -s tests -p test_parallel_search.py: three local tests passed.
• Validated live keyless discovery, web search and page fetch through the quickstart configuration and ToolManager. The orchestrator returned a sourced answer using scripted language model responses. Observed requests identified MiroFlow and carried no authorization header.

I work at Parallel. This contribution was developed with AI assistance.