A Model Context Protocol server that gives AI coding agents structured planning and sequential-thinking tools, so complex multi-file changes start from a reviewed plan.
Complex Plans is a Model Context Protocol (MCP) server that gives AI coding agents a structured way to plan before they act. Instead of diving straight into a complex, multi-file change, the agent writes a Markdown plan, asks you to review it, and only then starts implementing.
It also bundles a sequential thinking tool for step-by-step reasoning, so the same server covers both planning and analysis. It works with any MCP-compatible client: it was developed mainly for the Mistral Vibe CLI, but also connects to Claude Code and other editors.
npx -y @tuchsoft/mcp-complex-plansFor Claude Code, add the server to your configuration:
claude mcp add complex_plans npx -- -y @tuchsoft/mcp-complex-plans --disabled-tools=sequentialThinkingFor the Mistral Vibe CLI, add it to ~/.vibe/config.toml. The sequentialThinking tool can be disabled when the model already has native reasoning, to avoid redundancy.
createPlan - write a structured plan for a complex taskupdatePlan - edit an existing plan with search/replace blockslistPlans and readPlan - browse and load plansdeletePlan - remove a plan when it is doneopenInEditor - open the plan in your configured editorsequentialThinking - dynamic, multi-step reasoningBehaviour can be set through a project-level config file, CLI arguments or environment variables: the plans directory name, the default editor, whether plans are added to .gitignore, and which tools to disable. Everything operates safely inside the configured plans directory.