[FEATURE] strands-mcp: hydrate search_docs snippets concurrently

[yonib05]

SDK Language

Python

Problem Statement

search_docs hydrates up to 5 top results with sequential blocking HTTP fetches inside the tool call (SNIPPET_HYDRATE_MAX). First-search latency is up to 5x a single page fetch, and with the 30s default timeout a couple of slow or broken pages can make one tool call take minutes.

Proposed Solution

Fetch the hydration set concurrently (ThreadPoolExecutor is enough), or prefetch pages in a background thread at startup so searches never block on network. If the content-indexing fix (#3321) is implemented via startup prefetch, this may come for free.

Use Case

First search in a session returns in roughly one page-fetch time instead of five, keeping the MCP tool call responsive.

Alternatives Solutions

No response

Additional Context

Part of the strands-mcp improvements tracking issue #3317.

[March-77]

I’d like to take this. I’ll keep the change scoped to concurrent hydration of the bounded top-result set while preserving result ordering and response shape. I’ll add a deterministic regression test that proves independent page fetches overlap, exercise failure isolation, and report a synthetic fixed-latency before/after measurement alongside the MCP unit/lint checks.