Model Context Protocol
Connect a brain to
whatever you code in.
A brain speaks MCP, so any client that speaks MCP can read it. Below is the exact configuration for each one, taken from that client's own documentation.
Terminal
Editor
Desktop
Claude Code · run thisClaude Code docs ↗
claude mcp add --transport http mozg https://mozg.sh/mcp \ --header "Authorization: Bearer YOUR_TOKEN"
replace YOUR_TOKEN — sign in to make one
One command, nothing to edit. This is the path we test against.
Models, and where they fit
None of these connect to a brain directly. Each one runs inside a client from the list above.
Kimi K2MoonshotUse Kimi CLI above, or point Claude Code at Moonshot's endpoint.
Qwen3-CoderAlibabaUse Qwen Code above.
GLMZ.ai · ZhipuNo agent of its own. Run it inside Claude Code or Cline.
DeepSeekDeepSeekNo agent of its own. Run it inside Cline, Cursor or Claude Code.
MiniMax M2MiniMaxNo agent of its own. Run it inside a client from the list.
WanAlibabaVideo generation — it has no tools and no agent loop, so there is nothing for a brain to attach to.
What the agent gets
Five tools. The descriptions tell the agent when to reach for each one, which is the difference between a brain that gets used and one that sits there.
brain_listWhat brains am I allowed to read?brain_briefWhat does this brain cover — and what is it known to be missing?brain_searchFind what this project actually decided, before answering from general knowledge.brain_readOpen one note in full.brain_writeSave a convention or a pitfall worth keeping.