Patrick Buckley 664d44c109 fix(examples): rewrite mcp-cluster-ops to use console SDK for cluster… (#278) ...

* fix(examples): rewrite mcp-cluster-ops to use console SDK for cluster routing

The example MCP server was broken after the direct HTTP transport
refactor — it used TurnstoneServer (single-node) for cluster ops that
require TurnstoneConsole (cluster gateway). Rewrites dispatch flow to:
route via console → SSE stream from node → cleanup via console.

- Switch from TurnstoneServer to TurnstoneConsole for node listing and
  workstream routing (TURNSTONE_CONSOLE_URL replaces TURNSTONE_SERVER_URL)
- Add proper workstream lifecycle: create via routing proxy, stream from
  node, close in finally block with leak-safe ws_id guard
- Catch dispatch exceptions in run_on_node for structured JSON errors
- Extract _extract_node_ids helper, remove dead n.get("id") fallback
- Normalise _console_kwargs to always include token key
- Rewrite tests against Console+Server mocks (36 → 44 tests)

* fix(examples): paginate node listing and clarify auth in README

Address Copilot review feedback on #278:
- _list_nodes_sync now paginates via offset/limit loop so clusters
  with >100 nodes are fully discovered
- README step 2 now mentions token passthrough for authenticated clusters
- New test_paginates_large_clusters verifies multi-page fetch (45 tests)

2026-04-02 15:12:41 -07:00

..

mcp-cluster-ops

fix(examples): rewrite mcp-cluster-ops to use console SDK for cluster… (#278)

2026-04-02 15:12:41 -07:00