Alternatives
Compare the main Cursor-to-OpenCode integration approaches and their trust boundaries.
Several community projects connect Cursor models to OpenCode. They use different protocols, authentication methods, and tool-execution boundaries. Choose by mechanism, not by name.
| Project | Approach | Authentication | Main trade-off |
|---|---|---|---|
| open-cursor | Wraps cursor-agent, with an optional official Cursor SDK fallback | Existing CLI login or Cursor API key | Keeps the supported CLI path, but translates a process stream into an OpenCode provider |
| stablekernel/opencode-cursor | Official @cursor/sdk provider | Cursor API key | Direct SDK integration; its README warns that Cursor tools can run outside OpenCode permissions unless sandboxed |
| cursor-opencode-provider | Cursor Connect-RPC and protobuf provider | Browser OAuth or Cursor API key | Rich native protocol mapping with a larger compatibility surface when Cursor changes its protocol |
| yet-another-opencode-cursor-auth | Unofficial Cursor service interface | OAuth | Dynamic models and tool calls; the project labels itself experimental and warns about account and compatibility risk |
| opencode-cursor-auth | Local cursor-agent-backed OpenAI-compatible service | Cursor subscription and CLI login | Straightforward local service; its README calls tool calling experimental and notes missing thinking and usage accounting |
| cursor-opencode-auth | Standalone OpenAI-compatible proxy | Cursor CLI token from macOS Keychain | Works with any compatible client, but currently requires macOS and a separately running proxy |
The table reflects each project's own documentation as of July 2026. Check the linked repository before installing because activity, support, and protocol compatibility can change quickly.
Decision guide
Use this project when you already trust cursor-agent, want OpenCode's provider
surface, and prefer the CLI to remain the default backend.
Consider the official SDK approach when API-key authentication and direct SDK model discovery are more important than CLI reuse. Read its security boundary carefully before allowing file or shell tools.
Consider a standalone proxy when another OpenAI-compatible client also needs the endpoint. That adds a service to operate and protect.
Treat reverse-engineered service or RPC integrations as compatibility work. They can expose capabilities earlier, but Cursor can change an undocumented contract without preserving it.
Whichever route you choose, verify:
- where credentials are stored;
- which process can mutate the workspace;
- whether OpenCode permissions cover those mutations;
- how model IDs are discovered;
- whether the project documents its failure and account-risk boundaries.