Willder documentation
Connect your agents over MCP.
Start with the quickstart: one scope, one token, and a decision that reaches your next session.
Works with
Every client that connects uses the same endpoint. Each page below says how that client connects, or why it cannot yet. “Run end to end” means the client itself was run against Willder; “unrun” means its config is committed and the server answers it, but nobody has run that client yet. The verification table says how each status was checked.
- Claude CodeRun end to end
- Codex CLIRun end to end
- GooseRun end to end
- LangChain / LangGraphRun end to end
- OpenAI Agents SDKRun end to end
- Vercel AI SDKRun end to end
- Gemini CLIConfig committed, unrun
- Hermes (Nous)Config committed, unrun
- Claude DesktopConfig committed, unrun
- CursorConfig committed, unrun
- WindsurfConfig committed, unrun
- ContinueConfig committed, unrun
- VS Code (Copilot)Config committed, unrun
- Open WebUIConfig committed, unrun
- ZedConfig committed, unrun
- ClineConfig committed, unrun
- LM StudioConfig committed, unrun
- OpenClawConfig committed, unrun
- claude.ai (web)Not yet: needs OAuth
- ChatGPTNot yet: needs OAuth
- OllamaNot applicable: no MCP client
Willder is in early access. These guides cover the hosted MCP connection and its limits. New here? Start with the quickstart.