[SergeyMenshykh]

Motivation & Context

Honor server-provided TTLs for MCP-listed skills, avoiding repeated discovery while refreshing expired catalogs on demand.

Description & Review Guide

• Read ttlMs from skills/list responses and cache skills while every discovery page remains fresh.
• Refresh the complete catalog after expiration, preserving previously returned manifests and avoiding failed, canceled, or partial cache replacements.
• Document disabling additional caching layers with DisableCaching() and add deterministic TTL, pagination, concurrency, cancellation, and disposal tests.

Related Issue

N/A - no linked issue.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) - a workflow keeps the label and title prefix in sync automatically.

[Copilot]

🟢 Approval recommended

No blocking issues remain, and focused tests cover freshness, refresh failures, synchronization, cancellation, and disposal.

No open issues 🎉

What changed in this PR

Adds server-directed TTL caching to .NET MCP skill discovery, reducing repeated requests while refreshing expired catalogs on demand.

Changes:

• Parses per-page TTLs and synchronizes complete catalog refreshes.
• Integrates loader disposal and documents disabling outer caching layers.
• Adds deterministic coverage for expiration, pagination, concurrency, cancellation, and failed refreshes.

File	Description
dotnet/​tests/​Microsoft.Agents.AI.Mcp.UnitTests/​Skills/​Listed/​AgentMcpListedSkillsTests.Ttl.cs	Adds TTL and lifecycle tests.
dotnet/​tests/​Microsoft.Agents.AI.Mcp.UnitTests/​Skills/​Listed/​AgentMcpListedSkillsTests.cs	Makes tests partial and disposes loaders.
dotnet/​tests/​Microsoft.Agents.AI.Mcp.UnitTests/​Skills/​Listed/​AgentMcpListedSkillsSerializationTests.cs	Tests TTL deserialization and validation.
dotnet/​src/​Microsoft.Agents.AI.Mcp/​Skills/​Listed/​McpSkillsListResult.cs	Maps ttlMs to a nullable duration.
dotnet/​src/​Microsoft.Agents.AI.Mcp/​Skills/​Listed/​McpListedSkillsLoader.cs	Implements synchronized, TTL-aware catalog caching.
dotnet/​src/​Microsoft.Agents.AI.Mcp/​Skills/​AgentSkillsProviderBuilderMcpExtensions.cs	Documents disabling builder caching.
dotnet/​src/​Microsoft.Agents.AI.Mcp/​Skills/​AgentMcpSkillsSource.cs	Documents TTL behavior and disposes the loader.

_{Review effort: Balanced}

[github-actions]

MAF Automated Review — Iteration 1

Result: Findings reported
Scope: full PR (1 commit(s)): c29de4e2c6b0
Model: gpt-5.6-sol

Overview

The loader measures each page's TTL from response receipt, refreshes complete catalogs under a single-flight gate, and publishes replacements only after all pages and manifests validate. The tests cover expiry boundaries, pagination, cancellation, failed refreshes, and preservation of previously returned manifests. However, the default builder cache prevents the new TTL behavior from running at all unless every caller opts out, and the new disposal path can race active discovery.

Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
2 verified findings remained after source verification (1 high, 1 medium) across 2 files. Details are attached to the affected lines below.

Affected areas: dotnet/src/Microsoft.Agents.AI.Mcp/Skills/AgentSkillsProviderBuilderMcpExtensions.cs, dotnet/src/Microsoft.Agents.AI.Mcp/Skills/Listed/McpListedSkillsLoader.cs

[github-code-quality]

Code coverage overview

Languages: C#

C# / code-coverage/dotnet

feature-mcp-skills-n... 08851e6	sergeymenshykh-mcp-s... 834cc51	+/-
86%	85%	-1%

View coverage of the 10 most impacted files.

File	feature-mcp-skills-n... 08851e6	sergeymenshykh-mcp-s... 834cc51	+/-
/home/runner/wo...toryProvider.cs	95%	31%	-64%
/home/runner/wo...SessionStore.cs	100%	53%	-47%
/home/runner/wo...hellExecutor.cs	90%	73%	-17%
/home/runner/wo...SkillsSource.cs	89%	83%	-6%
/home/runner/wo...ctionVisitor.cs	81%	75%	-6%
/home/runner/wo...CopilotAgent.cs	72%	76%	+4%
/home/runner/wo...pListedSkill.cs	0%	93%	+93%
/home/runner/wo...SkillsLoader.cs	0%	100%	+100%
/home/runner/wo...dSkillsCache.cs	0%	100%	+100%
/home/runner/wo...ueExtensions.cs	0%	100%	+100%

ℹ️ Calculations represent line coverage. Results may change as calculations are updated.

---

_{Updated October 21, 2026 10:48 UTC}