Skip to content

Agentic development

Your AI agent deploys your app

A cloud API designed for AI agents. Give your coding assistant one scoped access code and it handles deployment, DNS, SSL and monitoring — while you keep building.
Get your agent codeExplore the Agent API

Free account in under two minutes — or skip signup with the one-line MCP setup.

Stay in flow

Deployment shouldn’t break your flow

You build an app with your AI assistant in minutes. Then the deploy eats the afternoon. Hand that part to the agent too.

One paste to production

Deploy without leaving your editor

No SDK to install and no dashboard to learn. Generate a scoped agent code, paste this into Cursor, Claude Code or Windsurf, and the agent discovers the platform and ships your app.
Get your agent code — freeHow access codes work
Paste into your AI assistant
You can deploy to Edge Network. Here's my agent access code:
ea_live_YOUR_CODE

Read https://edge.network/agent to discover what you can do, then deploy
this app: put static assets on Edge Storage, front them with the CDN,
add the DNS record for app.mysite.com, and issue SSL. Report the live
URL and the estimated monthly cost when you're done.

Official MCP server

One line of config. 22 tools.

Prefer MCP to raw HTTP? Add one entry to your tool’s config and Cursor, Windsurf, Claude Desktop or Continue gets native Edge tools — deploy a site, provision a VM, manage DNS, check project health.
.cursor/mcp.json
{
  "mcpServers": {
    "edge": { "command": "npx", "args": ["-y", "@edge-network/mcp"] }
  }
}
  • Zero-friction start. No code set? The server signs up for a free explore account on first run — no card, no dashboard visit.
  • Remote option. Hosted agents skip npx and POST to https://edge.network/mcp.
  • Framework wrappers. LangChain, LlamaIndex, CrewAI and Vercel AI SDK packages for agents you build.
Connect your toolTool reference

Designed for LLMs

Built for how agents think

Not a standard API with an agent wrapper bolted on. Every endpoint is shaped around how language models plan, act and explain.

  • Discovery-first

    One call returns the products, limits, existing projects, budget and a full self-describing schema. No documentation crawling.

    GET /agent

  • Budget-aware by default

    Every response carries cost context. Soft caps warn; hard caps stop new resources. Your agent works inside your limits.

    "cap_type": "soft"

  • Human-readable responses

    Every response includes tell_user — a plain-English summary your agent can relay to you word for word.

    "tell_user": "Live at …"

  • Dry run everything

    Preview any change and its cost before committing. Billable actions should always be dry-run first.

    X-Dry-Run: true

  • Project-based resources

    Group a VM, CDN, storage and DNS into one project. Health-check, scale or tear the whole app down in one call.

    GET /agent/projects/{id}/health

  • Errors as guidance

    When something fails, the error says what to try next and what to tell you — so the agent recovers instead of guessing.

    "suggestions": [ … ]

How it works

From access code to live app in five steps

  1. 01

    Create a free account

    Sign up in under two minutes. Free-tier deploys need no card.

  2. 02

    Generate a scoped agent code

    Choose the products, permissions and monthly budget it may use. Revoke it any time.

  3. 03

    Give it to your agent

    Add the MCP config to your editor, or hand any assistant the code and https://edge.network/agent.

  4. 04

    It discovers and deploys

    One call shows what it can do; then it creates storage, CDN, DNS and SSL and reports the URL.

  5. 05

    It monitors and reports

    Ask how things are going. It checks health, tracks spend against your cap and suggests fixes.

No lock-in

Works with any AI coding tool

The Agent API is a standard REST API. If it can make an HTTP request — an assistant, a pipeline, a script — it can deploy to Edge.

  • AI coding assistants

    Cursor, Windsurf, Cline, Claude Code, Continue — one-line MCP setup. Aider reads a conventions file.

  • CI/CD pipelines

    GitHub Actions, GitLab CI, Jenkins — put an agent code in your secrets and deploy from the pipeline.

  • Agents you build

    LangChain, LlamaIndex, CrewAI and Vercel AI SDK packages, the hosted MCP endpoint, or plain HTTP.

Discovery, three ways
# curl
curl https://edge.network/agent \
  -H "Authorization: Bearer ea_live_..."

# Python
requests.get("https://edge.network/agent",
  headers={"Authorization": "Bearer ea_live_..."})

# JavaScript
fetch("https://edge.network/agent", {
  headers: { Authorization: "Bearer ea_live_..." }
})

Or just tell your assistant: “Here’s my Edge agent code: ea_live_… Deploy my app to mysite.com.” Framework wrappers

You stay in control

Scoped, capped, logged and revocable

Agent access codes give you fine-grained control over what each agent can do. Set it once, or adjust it as the project grows.
  • Scoped permissions: pick the products and actions — deploy, manage, delete
  • Budget caps: soft caps warn, hard caps block new resources
  • Full audit trail: every action by every agent is logged
  • Instant revocation: revoke a code and access stops at once
Access code docs

Case study: Food & health

Peri Happy

A production stack, run by agents

Peri Happy is a nutrition platform for perimenopause, built by a tiny team with no dedicated ops. Their coding agent connects to Edge over MCP: it deploys the site to the CDN, manages DNS, runs the Compute instances behind their AI planner and configures Shield to keep bots off signup.

“We describe what we want and it’s live minutes later. The agent deploys, wires up DNS, and keeps watch — we spend our time on recipes and research, not infrastructure.”
Peri Happyfounding team
Read the Peri Happy case study
Ops headcount
0
Recipes served
552
Weekly readers
15k
Edge locations
988

FAQ

Before you hand over the keys

Can my agent run up a bill?
Only inside the budget you give its code. Soft caps warn the agent as it approaches the limit; hard caps block new resources once it’s reached. Billable tools support dry runs, so it can show you the cost first.
Do I need an account before connecting MCP?
No. Without an access code, the MCP server signs up for a free explore account on first run. It can deploy free-tier resources — static sites, DNS zones, storage buckets, Shield widgets — straight away. Anything billable stays dry-run only until you claim the account.
What does it cost?
Free-tier resources cost nothing, and egress is free on every product. Paid resources such as VMs are billed hourly from a prepaid balance. Once a card is on file you can claim a one-time $5 starter credit for Compute.
How is an agent code different from an API key?
API keys carry full account access. Agent access codes (ea_live_…) are scoped to chosen products and permissions, carry an optional budget cap, and can be revoked on their own without touching anything else.

Ready to let your agent build?

Create a free account, generate an agent code and start deploying in seconds.

Get your agent codeRead the agent docs

No card required for free-tier deploys. Budgets and hard caps mean nothing starts billing silently.