---
title: "Edge Agent API"
description: "The Edge Agent API lets AI agents deploy and manage compute, CDN, DNS and storage with discovery-first design, budget caps, dry runs and idempotent calls."
url: "https://edge.network/docs/agent"
section: "Agent tooling"
---

# Edge Agent API

Let AI agents deploy, manage, and monitor your infrastructure. Built for how LLMs think.

## How it works

You give your agent an access code and the URL, and it takes it from there:

1. **You** → "Here's my Edge agent code, deploy this app."
2. **Agent** → Discovers capabilities, deploys, reports back.
3. **You** → "How's it doing?"
4. **Agent** → Checks metrics, reports status, suggests optimisations.

Three principles make this work:

- **Discovery-first:** The agent learns what it can do, what limits exist and what's already deployed before taking any action.
- **Budget-aware:** Every response includes cost context. Set soft or hard budget caps so the agent operates within your limits.
- **`tell_user` field**: Every response includes a human-readable summary the agent can relay directly to you. No interpretation needed.

## Quick start

1. Go to [Account → Agent Access](/console/agent-access) and create an agent access code.
2. Give your AI agent the code and the URL: `https://edge.network/agent`.
3. The agent calls the discovery endpoint to learn what it can do.
4. The agent deploys, manages and monitors your infrastructure.

## Designed for agents

Unlike standard cloud APIs designed for humans or scripts, the Edge Agent API is built from the ground up for how LLMs work:

- **Dry run everything:** Preview any action before committing with the `X-Dry-Run: true` header.
- **Project-based:** Group resources by purpose (VM + CDN + DNS = one project). Health-check, scale or tear down everything at once.
- **Errors as guidance:** Error responses include actionable suggestions and a `tell_user` field explaining what the user should do.
- **Idempotent:** All mutating endpoints support idempotency keys for safe retries.

## Available products

| Product | What agents can manage |
|---|---|
| Compute | VMs, scaling, metrics |
| CDN | Deployments, domains, cache |
| Storage | Buckets, uploads, objects |
| DNS | Zones, records, propagation |
| Shield | Bot protection (free) |
| Assist | AI answers, grounded |

## OpenAPI specification

The full Agent API surface is published as [OpenAPI 3.0](/openapi.json), with typed schemas ready for function calling.

## In this section

- [MCP server](/docs/agent/mcp) — The official Edge Network MCP server, local via npx or hosted at edge.network/mcp: 22 tools for deploying and managing infrastructure from any MCP client.
- [MCP tool reference](/docs/agent/mcp-tools) — Every tool exposed by the Edge MCP server: all 22 tools with their parameters, types and descriptions, generated from the definitions the live endpoint serves.
- [Connect your tool](/docs/agent/connect) — Ready-to-paste Edge configs for Cursor, Windsurf, Claude Desktop, Grok Bot, Continue, Aider, LangChain, LlamaIndex, CrewAI and the Vercel AI SDK.
- [Framework wrappers](/docs/agent/frameworks) — Official Edge tool packages for LangChain, LlamaIndex, CrewAI and the Vercel AI SDK. Tool definitions are fetched from the MCP endpoint at runtime.
- [Agent self-signup](/docs/agent/self-signup) — AI agents can create their own Edge account autonomously, anchored to an Ed25519 keypair, then deploy free-tier resources until a human claims the account.
- [Agent access codes](/docs/agent/access-codes) — How agent access codes work: create scoped credentials for AI agents with product access, deploy or delete permissions, and soft or hard monthly budget caps.
- [Discovery endpoint](/docs/agent/discovery) — The agent's entry point to Edge Network: one call returns capabilities, limits, existing resources and a self-describing schema of every Agent API endpoint.
- [Deploy a static site](/docs/agent/deploy-static-site) — Deploy a static website with one Agent API call: storage, uploads, CDN, domain, SSL and DNS. Upload files directly or deploy from a GitHub or GitLab repo.
- [Deploy an application](/docs/agent/deploy-app) — Deploy a VM-based application with one Agent API call: provision compute, run a setup script, open firewall ports, attach SSH keys and add CDN and DNS.
- [Projects](/docs/agent/projects) — Group related resources into projects for unified management through the Agent API: one-call health checks, scaling, cost tracking and clean teardown.
- [Bot protection](/docs/agent/shield) — Let your AI agent add free Edge Shield bot protection to the forms and pages it deploys: create a widget, inject the embed and wire up server-side verification.
- [AI answers](/docs/agent/assist) — Let your AI agent add Edge Assist to the sites it deploys: grounded AI answers for visitors, created, indexed and embedded in two Agent API calls.