Skip to the content.

Koor Documentation

Lightweight coordination server for AI coding agents — “Redis for AI coding agents.”

Koor splits control (MCP, thin discovery layer) from data (REST + CLI, direct access), solving the MCP Token Tax Problem where coordination data gets routed through the LLM context window, burning tokens on data the LLM doesn’t need to reason about.

Doc Description
Getting Started Install, start the server, make your first API call in under 5 minutes
Configuration All server flags, environment variables, config file format, and priority rules
API Reference Complete REST API: every endpoint with method, path, request/response body, status codes, and ETag behaviour
CLI Reference Every koor-cli command with flags, examples, and expected output
MCP Guide Connect LLM agents via MCP. IDE config snippets for Claude Code, Cursor, and Kilo Code
Events Guide Pub/sub concepts, topic patterns, WebSocket subscriptions, event history, and use cases
Specs and Validation Shared specifications, validation rules (regex, missing, custom), filename filtering, worked examples
Deployment Local, LAN, and cloud deployment. Docker, systemd, Windows service, reverse proxy, and backup
Architecture Control/data plane split, technology choices, the “Redis for AI coding agents” concept, dependency rationale
Troubleshooting Common issues: auth errors, port conflicts, WebSocket problems, stale instances, build issues

Core Concepts

Shared layers:

  1. State Store — Key/value store with versioned history, rollback, and JSON diff
  2. Spec Registry — Per-project specifications with validation rules and compliance checking
  3. Event Bus — Pub/sub with SQLite history, WebSocket streaming, time-range queries, and webhook notifications
  4. Instance Registry — Agent registration, capabilities, liveness monitoring, and stale detection
  5. Audit & Observability — Immutable change log and per-agent metrics in hourly buckets

Additional features:

Two access methods:

Any LLM, any IDE:

Koor works with Claude Code, Cursor, Kilo Code, or any MCP-compatible client. The REST API works with any HTTP client, making it LLM-agnostic.