Sixty-three tools is a lot to put in front of a model, and the useful framing is the three levels the author describes: stateless code intelligence, embedded LLM agents, then distributed cluster work. The dependency to plan around is stated plainly in the README — without Redis, everything that remembers anything across sessions is off, and you are left with the code-intelligence and AI-call subset.
A large MCP server that extends a coding assistant with cross-session state, structured cognition and multi-agent orchestration. Redis holds the state between sessions; eight LLM providers can be called from inside it for work your editor's own model does not do. Works with Claude Code, Cursor, Windsurf, VS Code, Zed and any MCP client.
- 63 tools across 15 categories, with the consolidated ones taking an `action` parameter — `task action=create|get|list|update|delete`, for instance
- Code intelligence that works on meaning as well as text: `find-definition`, `find-references` and `semantic-search`
- Cognition tools that persist what a session learned — `decision-journal`, `error-pattern`, `confidence-tracker`, `intent-tracker`, `self-critique`, `smart-handoff` and `memory`, so the same Redis timeout bug is not re-diagnosed next month
- A Kanban layer agents drive directly: `task`, `task-multi`, `board`, `workspace`
- Multi-model review in one call — `multi-prompt` fans a prompt out to several models, `consensus-prompt` runs a board-and-CEO vote, and `mind-chain` hands a question through The Nine Minds (`socratic`, `ontologist`, `contrarian`, `simplifier` and the rest) so each sees the previous one's output
- Autonomous agent runs with `agent-orchestrate`, including 2–10 agents in parallel, each with its own orchestration ID; `monitor` reports status, and because MCP blocks during a tool call, live status is also available over HTTP at `/orchestrations`
- A cluster layer for distributed prompt execution — `cluster-bus-magistrale` dispatches one prompt across registered clusters under four strategies: first-wins, best-of-n, consensus, fallback-chain
No account for the server itself; API keys only for whichever model providers you use — `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`, `GROQ_API_KEY`, `DEEPSEEK_API_KEY`, `OPENROUTER_API_KEY` or `PERPLEXITY_API_KEY`; Ollama needs none. Bun installs it: `bun install -g kemdicode-mcp`, then `claude mcp add kemdicode-mcp -- kemdicode-mcp --stdio`. **Redis is required for persistence** — without it only the stateless tools (code intelligence, AI calls) function; `docker run -d -p 6379:6379 redis:alpine` is enough. HTTP transport for multi-session use runs with `--port 3100`. GPL-3.0 licensed.
One command plus a key — bun install -g kemdicode-mcp, then supply credentials