Entroly integrates through four distinct paths. A client name in this document does not imply that every subscription, OAuth flow, model, or provider route is intercepted.
Provider-bound token and cost measurements exist only when the request actually traverses the Entroly proxy. MCP-only integrations expose context selection, receipts, exact recovery, and verification tools, but do not automatically intercept every inference request made by the host.
| Agent or platform | Best Entroly path | Status | Authentication and routing boundary |
|---|---|---|---|
| Claude Code | Scoped MCP attachment; API-key proxy wrap | Native | Claude Pro/Max subscription sessions should use MCP. Public-API proxying requires ANTHROPIC_API_KEY. |
| Codex CLI | Scoped MCP attachment; API-key proxy wrap | Native | ChatGPT-account mode can bypass OPENAI_BASE_URL; provider-bound measurement requires API-key or custom-provider routing. |
| GitHub Copilot CLI | MCP for subscription sessions; custom-provider proxy for BYOK | Supported with mode boundary | MCP works with the signed-in CLI. Entroly does not claim interception of GitHub-hosted subscription inference. |
| OpenClaw | ContextEngine plugin and scoped MCP attachment | Native | OpenClaw retains provider authentication; Entroly assembles context and emits receipts. |
| Cursor | Automatic project MCP config; optional custom proxy endpoint | Automatic MCP | Restart Cursor after configuration. Proxy accounting exists only when the model route points through Entroly. |
| Aider | Session-scoped OpenAI-compatible proxy | One command | Requires an API/provider route that accepts a custom endpoint. |
| OpenCode | Session-scoped OpenAI-compatible proxy | One command | Provider authentication remains owned by OpenCode and its upstream. |
| Gemini CLI | Session-scoped Gemini-compatible proxy | One command | Requires GEMINI_API_KEY for the provider-bound proxy path. |
| Qwen Code | Session-scoped OpenAI-compatible proxy | One command | Confirm the selected provider honors OPENAI_BASE_URL. |
| Cline | Printed OpenAI-compatible endpoint settings | Guided setup | Entroly does not silently mutate extension settings whose schema may change by version. |
| Continue | Generated provider snippet | Guided setup | Confirm the active model entry uses the Entroly base URL. |
| Grok CLI | Custom model entry pointed at Entroly | Guided BYOK | Grok custom endpoints use API-key auth. Default signed-in inference is not claimed as intercepted. Entroly’s OpenAI-compatible upstream must be set to the intended xAI endpoint. |
| Goose | OpenAI-compatible endpoint | Compatible; end-to-end validation pending | Do not label one-command support until a watchdog-backed request test passes. |
| OpenHands | Local CLI LLM_BASE_URL route |
Compatible; end-to-end validation pending | A remote/cloud sandbox cannot reach a laptop-local proxy. |
| Mistral Vibe | Generic OpenAI-style provider in config.toml |
Guided setup | Use a project or user provider entry with the Entroly endpoint. |
| Oh My Pi | Custom provider in ~/.omp/agent/models.yml |
Guided setup | Stored OAuth credentials are not assumed to be valid for an arbitrary proxy. |
| ZCode | Custom OpenAI/Anthropic-compatible base URL | Guided setup | Confirm the chosen provider and authentication mode permit a custom URL. |
| Cortex Code | SDK/library boundary only | Not validated as a wrap target | No official, tested endpoint contract is currently documented by Entroly. |
Entroly supports two different Copilot CLI use cases and keeps them separate.
Use Entroly as an MCP server while Copilot continues to own its hosted model session:
entroly attach create --client copilot --project . --ttl 4h --install
This path can provide scoped context selection, Context Receipts, exact recovery, and verification tools. It does not claim that Copilot’s hosted inference request passed through the Entroly proxy.
Current Copilot CLI custom-provider settings use COPILOT_PROVIDER_BASE_URL, COPILOT_PROVIDER_TYPE, COPILOT_PROVIDER_API_KEY, and COPILOT_MODEL. Entroly uses the current copilot executable, sets the base URL and provider type, and leaves the provider key optional because local endpoints such as Ollama may not require authentication. COPILOT_MODEL is still required for custom-provider mode. This route is separately billed by the configured provider; it is not GitHub-hosted subscription inference.
Grok supports custom OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages endpoints in ~/.grok/config.toml. A custom model can point at Entroly, but the upstream must also be configured explicitly. Grok’s GROK_MODELS_BASE_URL mode fetches {base_url}/models; Entroly should not advertise that environment-variable path as one-command support until its model-list contract is tested end to end.
A client is proxy-compatible when all of the following are true:
Compatibility is not inferred merely because a client uses the word “OpenAI-compatible.”
A green status requires a code-backed setup path and a regression or end-to-end contract. Clients with documented custom endpoints but no Entroly request-flow test remain guided or validation pending. Hosted subscription interception is never inferred from MCP compatibility.
Preview a wrapper without changing configuration or launching the client with entroly wrap <agent> --dry-run; after a real proxy session, use the watchdog, receipt, or entroly value output to confirm that traffic actually traversed Entroly.