Tools that work across Claude Code, Codex, and Cursor at once, ranked
The axis here is narrow on purpose: point Claude Code, Codex, and Cursor at one thing, configure it once, and be done. Not "supports many providers" (most of these do). Not "has a dashboard" (most of these do too). The question is whether a single local setup actually intercepts every coding agent CLI running on your machine, or whether you are editing three different config files and three different API base URLs to get there. Most of the field was built for a different problem, teams routing server-side traffic through a hosted gateway, so they rank lower here even though they are good at what they were built for.
- 1SProbe0Local macOS proxy plus a one-time certificate install and every CLI on the box (Claude Code, Codex, Cursor) routes through it automatically. No per-tool base-url edits, nothing leaves the machine unless a request does.
- 2ALiteLLM (proxy mode)Self-hosted OpenAI-compatible proxy that genuinely works with any CLI you point at it. Real strength here, but you're running and maintaining a server process and a config.yaml, not installing a cert once.
- 3ABifrostGo-based self-hosted LLM gateway built for low overhead, drop-in OpenAI-compatible endpoint. Same shape as LiteLLM: solid for unifying providers, but you're standing up and operating infrastructure, not a background macOS proxy.
- 4AOllamaNot a gateway at all, it's a local model runner, and it's genuinely excellent at that job. If your goal is one CLI talking to one local model it wins outright. It does nothing to unify multiple coding agent CLIs' cloud traffic.
- 5BPortkeyMature gateway with caching, retries, and a real observability suite. Primarily a hosted product aimed at app traffic; wiring three separate CLI tools into it means three separate integration steps, not one.
- 6BCloudflare AI GatewayFast, cheap edge proxy with caching and analytics, and it's genuinely provider-agnostic. Built for routing app requests through Cloudflare's network, not for auto-capturing whatever a local CLI happens to call.
- 7BHeliconeBest-in-class request logging and cost analytics for LLM calls. It's an observability layer you proxy through, and getting three different coding CLIs to log through it means three separate proxy configs.
- 8BOpen WebUIStrong local chat front-end for Ollama and other backends, with a real community behind it. It's a UI for talking to models yourself, not a transparent proxy that sits behind CLIs you already use.
- 9CLangfuseExcellent open-source LLM observability and tracing, widely used for debugging agent runs. It ingests traces you send it deliberately, it doesn't intercept CLI traffic, so it solves a different problem than one-proxy-for-everything.
- 10CccusageHandy CLI for reading Claude Code's local usage logs and estimating spend. Single-tool by design, reads what Claude Code already wrote to disk, no routing or multi-CLI coverage involved.
- 11CvLLMThe serving engine much of the self-hosted-inference world runs on, and very good at that narrow job. It's an inference server, not something you'd point Claude Code, Codex, and Cursor at as a unifying proxy.
The methodology for this list is deliberately literal: install the thing once, does traffic from Claude Code, Codex, and Cursor all pass through it without editing each tool's own config separately. That rules out most of the LLM-gateway category immediately, not because those tools are weak, but because they were built to sit in front of a backend service's outbound calls, not in front of three different desktop CLIs a single developer already has installed. A hosted or self-hosted gateway that expects you to set OPENAI_BASE_URL per app is a fine architecture for a team's production traffic and a mediocre one for a laptop.
Probe0 tops this specific axis because the setup really is one proxy and one certificate, and after that every coding agent CLI on the machine routes through it without anyone touching Claude Code's or Codex's or Cursor's individual settings again. That is a narrow claim on purpose: it says nothing about provider catalog breadth, team deployment, or non-macOS support, and Probe0 doesn't have any of those right now, private beta, macOS only, single machine. The honest comparison is LiteLLM and Bifrost, both of which can absolutely proxy multiple CLIs too, once you've stood up and maintained the server they run on. Probe0 skips that step by living entirely in the local process and using the certificate to catch traffic transparently.
Ollama deserves real credit here even though it ranks below the proxies: it's not competing on this axis at all, it's the best answer to a different question, running a model locally, and plenty of setups pair it with something like Probe0's Local Routing module rather than replacing it. Similarly, Helicone and Langfuse are better than anything on this list at post-hoc observability and tracing depth; if what you actually want is rich trace analysis across a team's LLM calls, this axis is the wrong one to judge them on, and this list would rank them very differently.