Visual Studio Code 1.147 changes the default trust posture for MCP-delivered agent instructions. Microsoft’s release notes say Skills over MCP is now enabled by default, while first-use consent, catalog visibility, and a settings-based opt-out remain in place as of November 20, 2026.
For CTOs and engineering leaders, the signal is clear: MCP governance now needs to cover instructions as well as tools. This is not automatic trust in every server. It is automatic feature availability followed by a per-server approval decision.
1. Skills Over MCP Is Enabled by Default
What it is
Microsoft’s Visual Studio Code release notes state that Skills over MCP, based on SEP-2640, is enabled by default in VS Code 1.147. MCP servers can provide skills to the model without requiring developers to activate the feature first.
Microsoft’s Azure Dev Community announcement describes these skills as instruction packages that help models use tools and execute defined workflows. The announcement says compatible servers advertise the io.modelcontextprotocol/skills extension, while clients can discover available skills through skills/list and retrieve required content through resources/read.
This is a distribution mechanism for model instructions. It is not merely another tool endpoint.
The release notes preserve a user-consent boundary. Microsoft says VS Code asks for approval the first time a user connects to an MCP server that provides skills; after approval, the model can interact with those skills, and users can inspect them on the Skills tab of the Agent Customizations editor.
That sequence matters:
- Connect to an MCP server. VS Code detects that the server provides skills.
- Review the approval request. The user decides whether those skills can enter the model context.
- Inspect the resulting catalog. Approved skills appear in the Agent Customizations editor.
- Disable the capability if required. Setting
chat.agentHost.mcpSkills.enabledtofalseopts the editor out of MCP-provided skills.
Microsoft’s Community Hub announcement also says VS Code 1.147 and later can load agent skills from an MCP server. For platform teams, that version boundary turns editor rollout into a governance event rather than a routine client update.
Why it matters
The default has moved from activation to exception handling. Teams that permit MCP connections must now assume that a compatible server may offer model instructions as well as callable tools.
Why this matters: A tool defines an operation, while a skill can shape how the model approaches a workflow. Those controls belong in the same server review process because both affect agent behavior.
The approval prompt provides a boundary, but it places a decision in front of individual developers. For regulated or centrally managed environments, relying on inconsistent user judgments is not a complete control model.
Engineering leaders should establish an explicit posture before deploying VS Code 1.147 broadly:
- Allow: Keep
chat.agentHost.mcpSkills.enabledactive and limit connections to reviewed MCP servers. - Delay: Disable the setting while platform or security teams assess server-provided skills.
- Verify: Compare the Skills tab with the organization’s approved MCP server inventory after rollout.
- Document: Explain that approving an MCP server’s skills permits external instructions to enter the model context.
Microsoft says approved skills remain visible through the Agent Customizations editor. That visibility gives teams an operational checkpoint: developers can confirm what the model can use, while support teams can distinguish a skill supplied by an MCP server from other agent customization.
Perhaps the most consequential change is organizational rather than technical. A centrally maintained MCP server can distribute shared instructions, but a default-on client makes server ownership, change review, and approval guidance part of the developer-platform operating model.
The Bottom Line
Visual Studio Code 1.147 makes Skills over MCP available by default while retaining first-connection consent and the chat.agentHost.mcpSkills.enabled opt-out. The opportunity is consistent delivery of model workflows from MCP servers. The obligation is governing which servers may supply those workflows.
What This Means for Enterprise Teams
- Set the policy before rollout. Decide whether MCP-provided skills are permitted before distributing VS Code 1.147 across managed developer environments.
- Review instructions with tools. Treat a server’s skill catalog as part of the same assessment applied to its MCP tools and access.
- Standardize the setting. Define the expected value of
chat.agentHost.mcpSkills.enabledinstead of leaving the default to vary across teams. - Prepare developers for consent. Explain what the first-server approval means and which MCP servers are authorized.
- Audit through the product UI. Use the Skills tab in the Agent Customizations editor to verify the capabilities visible to the model.
Agent-first development is not only a model-selection decision. It is now a configuration and trust-management discipline.
Happy Coding!
