Kage MCP server

What it does

Kage is a memory management system for coding agents that stores decisions and context as verified Markdown files in your repository. It briefs agents with repo-specific knowledge, runs their work in isolated git worktrees, and independently verifies their claims against actual code before merging—keeping shared team memory accurate and up to date.

• Brief a coding agent with your repository's decision history and runbooks before delegating a coding task, so it starts with context about tricky deploys or known issues.
• Run checks yourself (tests, diff size, citation validation) on agent-generated code rather than trusting the agent's self-report, then merge only when all checks pass.
• Store and retrieve verified facts about your codebase as plain Markdown in your repo, shared with your team through git with no external accounts or databases required.

Tools this server exposes

Tool	Does
kage_room	Start an interactive session to brief and dispatch coding agents for delegated work.
kage_dispatch	Run a single delegated task with a specific agent, briefed from your repository's stored memory.
kage_runs	View the current status of all active agent runs.
kage_review	Inspect a completed run's verified claim, evidence logs, and code changes.
kage_merge	Land code from a run into your repository after verifying all checks pass, and record what the agent learned in your memory store.
kage_app	Start a local daemon and open a web UI to view runs, memory, and the room in a browser or native window.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.