Can Kimi K2.6 Run in Claude Code, Roo Code or Cline?

Claude Code is different. The Moonshot documentation that directly shows a Claude Code-style setup uses Kimi K2.5 through an Anthropic-compatible endpoint, not Kimi K2.6 .

Compatibility snapshot

Roo Code: OpenAI-compatible is only the first hurdle

Roo Code’s own documentation is explicit: it supports AI model providers that offer APIs compatible with the OpenAI API standard, but Roo Code uses native tool calling exclusively. If the selected model does not support OpenAI-compatible tool calling, it cannot be used with Roo Code .

That matters because Moonshot’s base_url is only the entry point. The Kimi K2 quickstart shows an OpenAI client configured with https://api.moonshot.ai/v1, but that alone proves the API route, not that Kimi K2.6 will behave correctly in Roo Code’s agent workflow .

A serious test should trigger file reading, file editing or another tool call. A single chat completion is not enough to prove Roo Code compatibility, because Roo Code’s requirement is about tool calling, not just text generation .

Cline: a plausible route, still short of direct proof

Cline documents an OpenAI Compatible provider configuration path, which makes Moonshot/Kimi-style OpenAI-compatible APIs a reasonable integration candidate . Kimi’s platform also has a K2.6 quickstart, and the Kimi K2 quickstart demonstrates OpenAI SDK usage with Moonshot’s base_url .

Still, those pieces do not equal a confirmed Kimi K2.6 + Cline integration. To upgrade the conclusion from “reasonable to test” to “confirmed,” you would need an official Cline or Kimi/Moonshot setup, or a reproducible third-party test using Kimi K2.6 specifically.

Claude Code: check Anthropic-compatible support, not just /v1

Claude Code should not be treated the same way as Roo Code or Cline. The cited Claude Code documentation here mainly covers model configuration and availableModels; it does not provide the same kind of OpenAI-compatible provider setup evidence that Cline and Roo Code provide .

The more relevant route for Claude Code is Anthropic Messages API compatibility. Ollama’s documentation says Anthropic Messages API compatibility makes it possible to use tools like Claude Code with Ollama models . Moonshot’s own programming-tools guide likewise shows a Claude Code-style setup for Kimi K2.5 using ANTHROPIC_BASE_URL=https://api.moonshot.ai/anthropic, ANTHROPIC_AUTH_TOKEN and the kimi-k2.5 model .

That is an important clue, but it is not proof for Kimi K2.6. A K2.5 Claude Code example only tells you which integration path to investigate; it does not prove that K2.6 is already supported through the same endpoint .

Pre-flight checklist before testing

1. Separate the endpoint types. For Roo Code and Cline, the obvious route is OpenAI-compatible provider configuration. For Claude Code, first verify whether Moonshot exposes Kimi K2.6 through an Anthropic-compatible path .
2. Do not guess the model ID. The Kimi K2 quickstart uses kimi-k2-turbo-preview; Kimi K2.6 should use whatever model ID is specified in the Kimi K2.6 quickstart .
3. Test tool calling, not just chat. Roo Code requires OpenAI-compatible native tool calling, so your test should include an agent task that actually invokes tools .
4. Record “connects” and “works as an agent” separately. A valid API key, base URL and model ID only prove that the connection can be made. Agent readiness still depends on tool calls, file operations and task completion.

Bottom line

If the question is “where should I try Kimi K2.6 first?”, Roo Code and Cline are the best candidates in this evidence set because they have OpenAI-compatible provider paths, and Moonshot/Kimi documentation shows OpenAI SDK / base_url usage .

If the question is “do public docs prove Kimi K2.6 works in Claude Code, Roo Code and Cline?”, the answer is no. Roo Code and Cline are plausible but not directly confirmed for Kimi K2.6. Claude Code still needs explicit K2.6 evidence on an Anthropic-compatible endpoint .