Current adoption status
| Claim | Status | Evidence boundary |
|---|---|---|
| UHP specification exists | Verified | Version 2026-08-11 is published in the official repository. |
| HarnessRouter Community Edition implements UHP | Verified | Official reference implementation; latest observed product tag is v0.8.3 (23 Aug 2026). |
| HarnessRouter Cloud implements UHP | Verified | Named by the UHP project as a hosted implementation. |
| Codex is supported as a HarnessRouter backend | Verified | Reference-implementation backend support. |
| Claude Code is supported as a HarnessRouter backend | Verified | Reference-implementation backend support. |
| Hermes is supported as a HarnessRouter backend | Verified | Reference-implementation backend support. |
| Pi is supported as a HarnessRouter backend | Verified | Reference-implementation backend integration (v0.7.0); not native Pi adoption. |
| DeepSeek Harness is supported as a HarnessRouter backend | Verified | Reference-implementation backend integration (v0.8.0); not native DeepSeek adoption. |
| mini-ork connects to UHP servers as an independent client transport | Verified implementation | Provider kind uhp: stdlib-only direct HTTP transport for UHP 2026-08-11 (/v1/responses, SSE, bearer auth, previous_response_id sidecar, example lane, real-server tests). Self-described "B1 of UHP adoption" — maturity in progress. Not a UHP server, not a conformance result. See UHP clients. |
| SuperQode consumes UHP servers as a third-party client | Verified | Independent client adoption: UHPClient/UHPHarnessProtocolAdapter and superqode connect uhp targets UHP 2026-08-11, with implementation code verified (discovery, tasks, streaming, continuation, cancellation, idempotency keys). This is not an independent UHP server, not a conformance result, and not vendor-native adoption. |
| Independent non-HarnessRouter UHP server | Not established | No separately maintained conformant server recorded by this tracker yet. |
| Native OpenAI/Codex UHP support | Not established | Backend compatibility through HarnessRouter is not native vendor adoption. |
| Native Anthropic/Claude Code UHP support | Not established | No primary vendor announcement naming UHP has been recorded here. |
| Native Hermes-project UHP support | Not established | HarnessRouter can run Hermes; that does not by itself prove the Hermes project implements UHP. |
| Native Pi UHP support | Not established | HarnessRouter can run Pi; that does not by itself prove the Pi project implements UHP. |
| Native DeepSeek Harness UHP support | Not established | HarnessRouter can run DeepSeek Harness; that does not by itself prove DeepSeek implements UHP. |
Implementation status vs adoption status
HarnessRouter's rapid release activity through v0.8.3 is evidence that the reference implementation is evolving. It is not evidence of a multi-vendor standard. The adoption threshold is intentionally higher: independent implementations, primary vendor announcements or cross-implementation conformance evidence.
What would upgrade a row?
- A vendor-native row becomes Verified only when a primary vendor source names UHP directly and documents support.
- An independent-server row requires a separately maintained implementation plus reproducible conformance evidence.
- An SDK/client row should demonstrate that it can target more than one UHP server without depending on private HarnessRouter behavior.
Why this tracker is conservative
Early protocol ecosystems are easy to overcount. Press-release syndication is not adoption, a backend adapter is not native vendor implementation, and a wrapper around the reference implementation is not automatically an independent server.
Last verified
. Re-check primary sources before making production interoperability commitments.