feat: add Tavily search as configurable MCP tool in miroflow-agent

[ParakhJaggi]

Summary

• Wired tool-tavily-search as a new configurable MCP tool in the miroflow-agent app, alongside the existing tool-google-search and tool-sogou-search providers (additive/parallel strategy).
• Added TAVILY_API_KEY environment variable loading in settings.py.
• Added tool-tavily-search block in create_mcp_server_parameters() that spawns miroflow_tools.mcp_servers.tavily_mcp_server with the Tavily API key.
• Added has_tavily_api_key flag to get_env_info() for observability parity.
• Documented TAVILY_API_KEY in .env.example.
• Added tool-tavily-search as a commented-out alternative in default.yaml.

Files changed

• apps/miroflow-agent/src/config/settings.py — env var, MCP server block, env info entry
• apps/miroflow-agent/.env.example — new TAVILY_API_KEY entry
• apps/miroflow-agent/conf/agent/default.yaml — commented-out tool option

Dependency changes

• None (no new Python packages added; the MCP server module miroflow_tools.mcp_servers.tavily_mcp_server is expected to be provided by a separate unit)

Environment variable changes

• Added TAVILY_API_KEY reference in settings.py and .env.example

Notes for reviewers

• This is an additive change — all existing search providers (tool-google-search, tool-sogou-search) and their dependencies are fully preserved.
• The referenced MCP server module miroflow_tools.mcp_servers.tavily_mcp_server does not yet exist in the repo. It needs to be created separately before tool-tavily-search can be activated.
• To enable Tavily search, set TAVILY_API_KEY in your .env file and replace tool-google-search with tool-tavily-search in your agent YAML config.

🤖 Generated with Claude Code

Automated Review

• Passed after 2 attempt(s)
• Final review: The Tavily migration is correct and complete. The new tavily_mcp_server.py follows the existing FastMCP pattern from serper_mcp_server.py, uses the correct TavilyClient SDK API, and is properly wired into settings.py with env-var validation, dependency declaration in pyproject.toml, and documentation in .env.example. No regressions — all changes are additive. Two minor quality issues exist (client instantiated per-call, no retry logic) but neither blocks functionality.