Changelog

All notable changes to perfmon-mcp are documented here. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Startup crash on the modern mcp SDK (#38). mcp SDK v2 removed the bundled mcp.server.fastmcp module (FastMCP now ships as the standalone fastmcp package), so a fresh resolve of the old unbounded mcp>=1.0 pin pulled mcp 2.x and crashed at import with ModuleNotFoundError: No module named 'mcp.server.fastmcp'. FastMCP is now imported from the standalone fastmcp package, fastmcp>=2 is an explicit dependency, and the mcp pin is a bounded mcp>=1,<3 range. Tool names and behavior are unchanged. Added tests/test_stdio_handshake.py (real JSON-RPC initialize + tools/list over stdio) plus a FastMCP-source guard in the smoke test to lock the migration.

get_rss_distribution(log_id, adapter_filter, scenario_hint) — RSS-specific lens scoped to the 13 curated Mellanox WinOF-2 RSS counter names; split into per-CPU + per-RqNum + per-SqNum sections with hot/idle counts in each section header.
discover_counter_instances(set_name, instance_filter, target) — enumerate per-instance counter paths for one PDH counter set via (Get-Counter -ListSet '<set>').PathsWithInstances; supports a regex filter (e.g. 'Adapter #2').
parse_counter_instances_output(text, set_name, instance_filter) — sibling parser for remote stdout.
get_teardown_commands(collector_name, target) — emit-only force-cleanup runbook for the managed collector + straggler perfmon / typeperf / relog / logman processes.
parse_teardown_output(text, collector_name) — sibling parser for remote teardown stdout.
compute_rate_from_counter(log_id, counter_filter, interval_s) — per-counter rate aggregator for monotonic raw totals: (last - first) / DurationSeconds.
get_per_queue_summary now surfaces Delta, MaxMinRatio, Hot, and Idle peer-group flags plus a footer summary.
instance_filter keyword argument on get_capture_commands / get_capture_instructions, narrowing per-instance counter enumeration.
Mellanox NIC RSS workflow absorbed the standalone analyze-mellanox-rss PowerShell skill into a discoverable MCP surface so the LLM can drive every step through tool calls without leaving the model loop.

Profile metadata now carries priority_metrics and default_instance_filter. mellanox-percpu defaults instance_filter to 'Adapter #2'.

analyze(log_id, sections) mega-tool that composes summary + timeline + per-queue overview into one call.
Log registry tools: load_log, load_csv, unload_log, log_info, list_loaded_logs, list_blgs.
Discovery tools: discover_counter_sets (vendor-tagged), discover_nics, get_capture_status (managed collector query).
get_counter_throughput — NIC throughput convenience lens narrowing the per-counter summary to Network Adapter throughput rows.
LabLink-first remote runbooks (with JSON sidecars carrying parse_with, shell, expected_runtime_s, timeout_s) across every emit-style tool.

load_blg — preserved as an alias of load_log for v0.2 compatibility; emits a DeprecationWarning.