Why do you guys keep changing the way MCP servers are called. For example. I use an app called tynn.ai to handle my project scope. I had a entry in the allow list for tynn:* and that was working perfectly for all projects. Now every project I have is asking permission to run the mcp tools and it’s using a funky syntax like project-0_app-tynn:next. This should just be tynn:next
What’s going on? I’m seeing this same issue with other mcp servers as well.
Ok now a funny thing keeps happening. In a single chat using Opus4.5, the agent used the mcp tool calls like that showed in the OP. I switched to auto to have it create the individual work in tynn and it changed the format it uses. I clicked the Allowlist MCP Tool option AND I added this format with wildcard to the allow list but it is asking me for permission on every single attempt. I’m loading up a new version in my roadmap that will have over 14 stories and nearly 100 tasks. This is so annoying.
I’ve been investigating this issue and found that the server identifier format changes based on how you open the project:
Single folder workspace: project-{N}-{server_name}
Example: project-0-pw-git
Multi-root workspace (when you add directories to workspace): {folder_name}-{server_name}
Example: project-pw-git
So the project-0 prefix disappears when using multi-root workspaces, but the core problem remains — each project folder still gets its own unique identifier due to the folder name being included in multi-root mode.
this issue is hit and miss. Sometimes I open a project and it requests using the project-0 pattern. another day I will open the same single folder project and it is using the correct pattern without this project-0 nonsense. I have even had this happen within the same chat but when changing models.