Unclear It is not clear whether any tool calls in this release are actually marked as coming from an MCP skill.
What
MCP (Model Context Protocol) servers are outside programs that give Claude extra tools and skills. When a tool call that comes from an MCP skill needs your permission and the request is passed to you remotely, Claude Code now handles it differently:
- The prompt describes the action with the skill's summary instead of the tool's standard description.
- The prompt does not offer an option to always allow the action.
- Channel servers are connected servers that pass permission prompts on to you somewhere else. The server that owns the skill no longer receives the prompt.
- An "allow" answer from the server that owns the skill is ignored.
Before, every permission prompt used the tool's standard description, and every connected channel server received it.
Why
This changes who can approve an MCP skill's actions. The server that provides a skill can no longer approve that skill's own requests.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear whether any tool calls in this release are actually marked as coming from an MCP skill.