Delega — Stop AI Agents Starting From Zero | MCP + API + CLI

# Delega > Stop making your AI agents start from zero. Delega gives coding agents shared task memory, durable context, and decision history across sessions, terminals, and tools. For the full documentation in one file, see: https://delega.dev/llms-full.txt ## If you are an AI agent Start here: https://delega.dev/agent Read the skill file: https://delega.dev/skill.md Machine-readable discovery: https://delega.dev/.well-known/agent.json OpenAPI contract: https://api.delega.dev/v1/openapi.json The agent onboarding flow gives you a restricted API key and sandbox project with one POST request. No human approval needed to start working. Your human can claim the account later. ## What is Delega? Delega is persistent task and decision memory for AI coding agents. Agents resume across sessions by reading their open tasks and durable per-task context, track execution and decisions in an audit trail, and can delegate to other agents when needed. Humans can inspect and steer that work in the browser task console — all with a safe hosted onboarding flow. ## Hosted onboarding for agents - Agent page: https://delega.dev/agent - Skill file: https://delega.dev/skill.md - Discovery JSON: https://delega.dev/.well-known/agent.json - OpenAPI contract: https://api.delega.dev/v1/openapi.json - Hosted signup endpoint: https://api.delega.dev/v1/agent/signup - Claim resend endpoint: https://api.delega.dev/v1/agent/claim/request - OTP verify endpoint: https://api.delega.dev/v1/agent/claim/verify - Capabilities endpoint: https://api.delega.dev/v1/agent/me/capabilities ### Hosted restricted-mode summary - Hosted tier only in V1. - One sandbox project per pending agent. - Restricted keys can create, list, update, complete, and comment only on tasks they created in their sandbox. - No pre-claim delegation, webhooks, or project creation. - Human approval happens through the human-controlled email, not the agent inbox. - The sandbox remains visible after claim. ## Key product features