fix(mcp_client): enforce overall timeout for list_tools pagination (#836)

[Artemon-line]

Summary

Enforces a top-level tokio::time::timeout wrapper across the complete list_tools and call_tool client lifecycle (DNS resolution, transport connection/handshake, and multi-page tools listing pagination).

Previously, each individual tools/list page request had its own timeout check inside paginate_tools. A degraded or malicious MCP server returning paginated tools (up to 100 pages) could delay every page request by slightly under the timeout duration, causing list_tools to execute for up to ~100x the configured timeout.

Related issue

Closes #836

Validation

• Unit tests (cargo test -p praxis-ai-apis -- mcp_client)
• Integration or functional tests (added list_tools_cumulative_pagination_timeout with mock slow paginated MCP server)
• make lint

Checklist

• I reviewed every changed line and can explain the change.
• New capabilities include an example config and functional example test.
• User-facing behavior and generated documentation are updated.
• Performance-sensitive changes include appropriate benchmark or load-test evidence.
• Commits are signed and include a Signed-off-by trailer.

Breaking changes

None.

[praxis-bot]

MCP Client Cumulative Timeout Review

Purpose: Wraps list_tools and call_tool in a top-level tokio::time::timeout so that cumulative pagination, DNS, and transport time cannot exceed the configured timeout. Previously a malicious MCP server could delay each paginated page by just under the per-page timeout, causing list_tools to run for up to ~100x the configured duration.

Assessment: The outer timeout correctly closes the cumulative-pagination DoS vector described in #836. The new test (list_tools_cumulative_pagination_timeout) exercises the fix with a real mock MCP server. The parse_display_url extraction is a clean way to make a display URL available for the outer timeout error without running resolve_and_validate first.

Severity	Count
Critical	0
Large	0
Medium	1

[praxis-bot]

Medium: inner timeouts are now unreachable dead code.

The outer tokio::time::timeout(timeout, Box::pin(work)) at line 256 uses the same timeout duration as every inner timeout: the connection timeout here (line 243), the per-page timeouts inside paginate_tools (line 353 in the unchanged code), and the call_tool timeout in the other function (line 316). Because the outer timer starts strictly before any inner timer using the same duration, the inner timeouts can never fire -- the outer one always expires first and cancels the inner future.

This means ~30 lines of inner timeout/map_err handling across both functions, plus the timeout parameter on paginate_tools, are dead code paths that will never execute.

Suggested change: either remove the inner timeouts (simplest) or, if per-operation limits are desired alongside the overall limit, use a shorter per-operation duration (e.g. timeout / 2 or a separate config field). Keeping both at the same duration is misleading -- a reader would assume both can fire.