TL;DR — A project can no longer replace one of your own MCP servers, a failed MCP server now says why, and /mcp browse drops the servers that are no longer maintained.
Security
- A project's MCP server can no longer replace one of your own. When
.codeep/mcp_servers.json or .mcp.json names a server the same as one in your global config, your server runs — even in a trusted workspace — and a one-line notice names the project entry to rename if you want both.
Fixed
- A failed MCP server says why. "exited (code 1)" now carries the last
lines the server printed (npm's noise dropped, environment values masked). A missing uvx or npx says what to install. A server that keeps crashing shows its last reason in /mcp.
/mcp browseno longer offers servers that are no longer maintained:
GitHub, GitLab, Postgres and Slack (deprecated on npm) and SQLite (archived on PyPI). Brave Search now runs Brave's own @brave/brave-search-mcp-server. If you installed one of them, remove it with /mcp remove <name>.
/mcp installrefuses an entry that is missing a required argument
(for example filesystem without a directory) instead of saving a server that cannot start; /mcp install git shows its repository as the optional --repository <path> the server actually takes.