# CodeOrch > CodeOrch is a persistent runtime for AI coding agents. It gives Claude Code, Cursor, Aider, and any MCP-compatible agent a shared memory layer and knowledge graph so every task compounds instead of starting from zero each session. CodeOrch connects to your GitHub repo, indexes your codebase into a hybrid store (PostgreSQL + pgvector), and exposes 28 typed MCP tools that any agent can call to receive a surgical 2–6k token briefing instead of reading tens of thousands of tokens of raw source. ## Product Pages - [Home](https://codeorch.app/): Product overview, manifesto, pipeline, and pricing summary - [Story](https://codeorch.app/story): Step-by-step explanation of how CodeOrch works — the amnesia problem, the graph, and the briefing pipeline - [MCP Tool Reference](https://codeorch.app/mcp): Full documentation for all 28 MCP tools available to agents - [Pricing](https://codeorch.app/pricing): Free, Pro ($9/mo), and Business ($19/mo) plans - [Use Cases](https://codeorch.app/use-cases): How teams use CodeOrch across different workflows - [FAQ](https://codeorch.app/faq): Common questions about setup, pricing, and agent compatibility ## For AI Agents - [agents.md](https://codeorch.app/agents.md): Plain-text listing of all 28 MCP tools with descriptions and call signatures — machine-readable - [MCP Docs](https://codeorch.app/mcp): Human-readable MCP tool reference with examples ## Authentication & API - Sign up: https://codeorch.app/signup - API base: https://api.codeorch.app - Auth: Bearer token (JWT), obtained after login - MCP server available post-signup; connection instructions in dashboard ## Key Capabilities - 70%+ token savings via orchestrated briefings (2–6k tokens vs 15–30k raw reads) - Persistent hybrid memory: PostgreSQL for tasks/decisions, pgvector for semantic search - GSAP-like scrub of task history so agents skip work already done - Pre-check greps that prevent re-reading files not changed since last index - Automatic PR creation on review gate pass - GitHub integration with encrypted proxy - Skills marketplace for domain-specific orchestration rules ## Supported Agents Claude Code, Cursor, Aider, and any MCP-protocol-compatible agent.