Providers and extensions
Connect your own AI provider and understand where tools run.
Open Settings → Local Server → Providers to configure AI providers for local work. Available authentication methods depend on the provider: some offer API keys, others a sign-in flow. Models are listed separately under Models.
Your providers in a shared project
Select the shared server in Settings, then open Providers. This page manages your credentials on your computer for that shared server. It does not reveal or edit the host's provider accounts, and it does not reuse an unrelated local-server provider profile.
When you send a shared request, the host runs the project agent and tools, while model inference is routed through your local provider runtime. The host runner does not receive your provider API key or OAuth token. The conversation and tool context needed for the request are sent to your selected provider.
Your local provider catalog and connection settings can load while the host is offline. Signing in may still require an internet connection to the provider. Running the shared project requires the host to return.
Provider usage is subject to that provider's pricing, quotas, and account terms. Worktree does not include model credits or sell access to another participant's provider subscription.
Extensions, MCP, and plugins
Local server and project extension settings expose the OpenCode backend's supported integrations. Tool execution follows the project's environment and permissions. A local MCP process or plugin used by a hosted project runs on the host, not automatically on every guest's computer.
Only install and configure executable extensions you trust. A plugin, MCP command, startup script, or dev-server reload can run with the host's privileges. Shared guests do not get host-wide provider or extension administration.
See Skills for reusable instructions and Permissions for the current execution restrictions.