Point it at your Ollama, vLLM, llama.cpp, or LM Studio server. Real checks from DNS on down — plus copy-paste config snippets to wire your agents.
Honest limits: checks run from ClawCurrent's server, not from your agent host — a PASS means the endpoint is reachable and well-formed from the public internet, not that your specific machine can reach it. Server identification is best-effort. Anything we cannot verify is marked UNKNOWN, never guessed. This is a free acquisition tool — rate-limited per IP, no account required.
Before you blame the agent framework, prove the endpoint. Each layer is checked independently and reported PASS / WARN / FAIL / UNKNOWN with evidence.
Hostname resolution, port reachability with latency, and TLS certificate validity dates for https endpoints.
GET /v1/models (OpenAI-compatible: Ollama ≥0.1.24, vLLM, llama.cpp server, LM Studio, text-generation-webui), GET /api/tags (Ollama native), plus /health and /metrics presence.
Best-effort identification from response shapes and headers, with the served model list (names and sizes when the server reports them).
Server-side harnesses don't care about CORS — browser-based agents do. We check the headers so you know which world you're in.
Copy-paste OpenAI-compatible base_url snippets and generic OpenClaw/Hermes-style config blocks pointing at your verified endpoint.
http/https only, DNS re-resolved and re-validated on every redirect hop, private/loopback/reserved ranges refused. The doctor cannot scan internal networks — including ours.