Choose an MCP Client¶
Complete the prerequisites and Unity package installation, choose the matching client setup below, then return to first-connection verification.
| Client | Primary configuration | In-Unity Chat |
|---|---|---|
| Claude Code | Project .mcp.json | Claude backend |
| Claude Desktop | Wizard clipboard or global CLI | Not used |
| Codex | Project .codex/config.toml | Codex backend |
| Cursor | Project .cursor/mcp.json | Not used |
| Windsurf | Global legacy Cascade mcp_config.json | Not used |
| VS Code | Project .vscode/mcp.json | Not used |
| Kimi | Global CLI | Kimi backend |
| OpenCode | Global CLI | OpenCode backend |
| Junie | Project .junie/mcp/mcp.json | Not used |
| Rider AI Assistant | Rider Settings UI | Not used |
| Gemini | Deprecated | Not supported |
Project-local configuration stays with one Unity project. For global configuration, use the canonical Python CLI command with the client key from the matching guide.
Do not copy configuration between clients: Codex uses TOML, VS Code uses a servers object, OpenCode uses its own local-server shape, and the supported JSON clients use mcpServers. Windsurf's generated project file is only a compatibility artifact; use the global path documented in its guide for legacy Cascade.
Optional Project Guidance¶
After configuring Claude Code or Codex, install the bundled AI skills and agents when you want project-local MCP workflows and focused Unity subagents. This installation is separate from the MCP server configuration above.