Setting up

Kimi, GLM, and other models

Use ContextStream with Kimi, GLM, Qwen, DeepSeek, MiniMax, or Muse Spark by setting it up for the coding tool that runs the model.

The short answer

ContextStream runs inside your coding tool, so it works with whichever model that tool uses. Set it up for the tool, not the model:

You runSetup
Kimi Code CLIcontextstream-mcp setup --editors=kimi
ZCode, Z.ai's GLM agentcontextstream-mcp setup --editors=zcode
Qwen Codecontextstream-mcp setup --editors=qwen
Muse Code on Muse Sparkcontextstream-mcp setup --editors=muse
Claude Code on Kimi or GLMcontextstream-mcp setup --editors=claude
OpenCode or Crush on Kimi, GLM, Qwen, DeepSeek, or MiniMaxcontextstream-mcp setup --editors=opencode or --editors=crush

Each tool's exact files are in the editor picker.

Claude Code on Kimi or GLM

Point Claude Code at the provider's Anthropic-compatible endpoint and put the provider's key in ANTHROPIC_AUTH_TOKEN:

ModelANTHROPIC_BASE_URL
Kimi (Moonshot)https://api.moonshot.ai/anthropic
GLM (Z.ai)https://api.z.ai/api/anthropic

ContextStream's tools, instruction file, and hooks work the same on either endpoint.

Tool definitions

ContextStream's tool definitions follow the strictest schema rules among these providers. Moonshot, for one, rejects a request when any tool uses a keyword outside its dialect. The same definitions work on Claude, GPT, Gemini, Kimi, GLM, and Qwen, so switching models never changes which ContextStream tools you have.

If it doesn't work

--editors=kimi (or another new id) is rejected. Your ContextStream binary predates that client. Run contextstream-mcp update, then run setup again.

The provider rejects the request with a schema error. Update the binary; older releases sent tool definitions some providers refuse. Then restart the coding tool so it reloads the tool list.

Claude Code still answers from Anthropic models. It reads ANTHROPIC_BASE_URL at startup. Set it in the shell that launches Claude Code, then start a new session.

Next