// AI-native deployment & runtime infrastructure

Give your agents autonomy. Stay in control.

Your agent decides what to build and how. AgentDeploy coordinates deployment, provisioning, connectivity and operations across infrastructure you choose and control.

Deploy and provision apps and services across your infrastructure—from any supported agent, anywhere.

  • Infrastructure stays under your control
  • Scoped, temporary agent access
  • Preview before production
MCP https://mcp.agent-deploy.ai/mcp
agentdeploy / example session
  1. > Publish this site as a preview.
  2. Build complete. Preparing release.
  3. ↳ get_static_packaging_guide
  4. ↳ create_static_upload
  5. ↳ complete_static_upload
  6. ↳ deploy_preview preview
  7. ✓ preview-r7.site.example
  8. > Looks good. Ship it to production.
  9. ↳ promote_preview release r-7
  10. ✓ Live. Same release. No rebuild.
release readyr-7 / sha256:9f2c…e41a
Illustrative session · real MCP tools · example URLs
Works with
  • Claude Code
  • OpenAI Codex
  • Cursor
  • GitHub Copilot
  • Gemini CLI
  • OpenCode
  • more MCP clients

// One neutral control plane

Bring the infrastructure you already choose under one operating model.

AgentDeploy connects your agents to infrastructure you choose and control. It coordinates deployment, provisioning, connectivity and ongoing operations without selling or operating hosting capacity.

Coding & general-purpose agentsClaude Code / Codex / Cursor / Copilot / MCPBuild artifact + deployment intent
AgentDeployControl plane
01 / DEPLOY

Deploy & publish

Verified releases, chosen environments, health checks and service entry points.

02 / CONNECT

Connect within boundaries

Gateway, Mesh and Tunnel paths with explicit identity and access policies.

03 / OPERATE

Operate over time

Observed status, logs, stop and resume, version rollback and audit records.

Customer-controlled infrastructure

Run in your cloud accounts, provider runtimes, edge sites or private data centers.

AWS / GCP / Azure / Cloudflare / Edge / On-prem
Your existing systems

Keep identity, internal APIs, CI/CD and observability in your operating model.

Explore enterprise deployment ↗
Open customer-side software

The Edge Runtime and customer-side clients will be open sourced so teams can inspect behavior, trust boundaries and updates.

Edge Runtime / clients / adapters

// How it works

From your workspace to your users.

  1. 01

    Connect once

    Add the hosted MCP endpoint and sign in. You choose which organizations the agent may touch; access is checked again on every request.

  2. 02

    Build locally

    The agent builds with its own tools in your workspace. AgentDeploy never needs your repository or a shell on your machine.

  3. 03

    Upload a release

    The built artifact is uploaded and verified as an immutable release. Your agent can inspect its status at every step.

  4. 04

    Preview, then promote

    Validate on a separate preview, then promote that exact release. Production always requires your explicit confirmation.

// Autonomous within your boundaries

Let agents act. Keep access and production under control.

website / releasesIllustration
⌘r-7verified artifact
↗previewseparate environment
◉productionr-6
deploy_preview✓ ready preview-r7.site.example

Production stays on r-6 while you review r-7.

Access stays inside your boundaries

Choose the organizations, infrastructure and resources your agents may reach. Every request is checked against current permissions.

Authority can expire with the task

Use scoped network tickets and temporary policies so access can end when work completes or the grant expires.

Production remains an explicit decision

Agents can prepare and verify a release. Promotion to production still requires your approval and respects the latest observed state.

// Runtimes

Three workload profiles for infrastructure you control.

static

Static sites

HTML or any built frontend, published as a verified artifact to a compatible target you control.

dist/
├── index.html
├── styles.css
└── assets/

Websites · landing pages · frontends

worker

Workers

TypeScript fetch handlers and bound resources, deployed to compatible customer-controlled or third-party runtimes.

export default {
  async fetch(request, env) {
    return new Response("live");
  }
}

APIs · webhooks · data-backed apps

oci

Containers

Pinned Linux OCI images serving HTTP, placed on eligible nodes in your cloud, edge or private infrastructure.

image    app@sha256:…
port     8080
health   /health
runtime  linux / oci

Existing services · custom stacks

// Connect

Let your agent help you connect.

Copy this prompt into your agent. It will follow the official guide to help configure MCP and walk you through sign-in.

Ask your agent

MCP setup prompt

Help me connect this agent client to AgentDeploy. Read https://install.agent-deploy.ai/agent first, identify my client and operating system, and check its existing MCP configuration. Add https://mcp.agent-deploy.ai/mcp using the supported setup for this client, preserving other servers. Install the official publishing Skill if supported. Guide me through browser sign-in and organization authorization. Then verify the MCP connection and report the organizations I can access. If you cannot change the configuration directly, give me the exact steps for this client.

Or set up manually

# add the server, then run /mcp to sign in
claude mcp add --transport http --scope user agentdeploy https://mcp.agent-deploy.ai/mcp
# install the publishing Skill
curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client claude

Gemini CLI, GitHub Copilot, VS Code, ChatGPT, stdio and more →

Then try
“Publish this project to AgentDeploy as a preview and send me the link.”

// Enterprise deployment

Teams choose the tools. Your organization sets the boundaries.

Business and engineering teams use approved agents to build applications. Platform teams bring existing cloud, edge and private infrastructure under one control plane and define where workloads may run.

01

Prepare the environment

Define allowed nodes, environments, network paths and user access before teams deploy.

02

Connect the application

Publish a controlled service entry and provide an authorized Gateway path to the internal APIs it needs.

03

Keep operations accountable

Inspect deployment, runtime and entry status. Associate versions and actions with the responsible identities.

A practical first pilot

One internal application, one test environment, one Gateway, one internal API and one user group. Verify deployment, authorized access and a complete rollback.

Explore the documentation ↗

OAuth, open APIs and runtime metrics are available. Enterprise SSO/SCIM, approval workflows and external system adapters are scoped to the integration project.

// A few useful answers

Before your first deploy.

Do I need to change how my agent builds?

Keep your existing editor and build tools. Your agent builds locally, then uses AgentDeploy MCP tools to package and publish the output.

Which agents can connect?

Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode and other clients with remote MCP support. The agent guide has the current setup options.

Where does my application run?

In infrastructure you choose and control: your cloud accounts, third-party runtime providers, edge sites or private data centers. AgentDeploy coordinates eligible targets through the control plane.

Can the agent publish straight to production?

The recommended flow is a separate preview followed by explicit approval to promote that exact release. Your agent can then inspect status, read logs and roll back supported deployments.

How does this fit our existing operations?

Your teams and infrastructure providers keep responsibility for capacity, application quality and data permissions. AgentDeploy unifies delivery, connectivity and agent automation with APIs for your existing systems.

Does AgentDeploy provide hosting capacity?

No. AgentDeploy is a neutral third-party control plane and does not sell or operate hosting capacity. The customer-side Edge Runtime and clients will be open sourced for transparency.

Let your agents act. Keep infrastructure yours.

Connect an agent, unify the infrastructure you already choose and automate operations within boundaries you control.

> built for agents. ready to deploy.