AgentDeploy Agent Tools

// Application publishing for coding agents

Build locally.Publish through MCP.

Publish and manage applications with your agents.

Choose how your agent connects

Hosted MCP serves remote agents, stdio handles local packaging, and the Skill supplies the reusable publishing workflow.

01 / HOSTED Recommended

Hosted MCP

Connect an OAuth-capable MCP client. Access is limited to the organizations you select and evaluated again on every request.

Streamable HTTPOAuth 2.1 + PKCENo install
https://mcp.agent-deploy.ai/mcp

Add this URL as a remote MCP server, sign in to AgentDeploy, then choose the organizations the client may access.

02 / LOCAL

stdio MCP

Run the compact local binary when your client does not support hosted OAuth or the agent must package a local build directory directly.

LinuxmacOSWindowsamd64 + arm64
curl -fsSL https://install.agent-deploy.ai/mcp | sh

Configure an organization UUID, private API-token file, and optional allowed workspace in the MCP client's process environment.

03 / WORKFLOW

Agent Skill

Install reusable instructions so a coding agent knows when to build, which MCP tools to call, how to upload, and when a deployment is ready.

SKILL.mdStatic + Worker packagersReusable
curl -fsSL https://install.agent-deploy.ai/skill/publish | sh

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

Develop, preview, promote

Build locally, validate on a separate Preview, then promote the tested immutable Release with explicit production authorization.

Static HTML / built frontend

Publication integrated

static

TypeScript Fetch Worker

Publication integrated

worker

OCI HTTP container

Publication integrated

oci

Next.js / vinext (pinned candidate)

Locally accepted candidate; not released

worker

JavaScript Worker + WebAssembly module

Publication integrated

worker

Python frameworks (runtime evidence only)

Runtime only; publication unavailable

worker

Read the live stack catalog

Connect the agent you already use

Follow a verified setup path for leading coding and office agents. Each guide uses the shared AgentDeploy MCP endpoint and canonical publishing Skill.

Coding agents

OpenAI CodexHosted OAuthAgent Skill
01 / Add the MCP server
codex mcp add agentdeploy --url https://mcp.agent-deploy.ai/mcp
02 / Authorize AgentDeploy
codex mcp login agentdeploy
03 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client codex
04 / Verify the connection
codex mcp list
Official client documentation ↗
Claude CodeHosted OAuthAgent Skill
01 / Add the MCP server
claude mcp add --transport http --scope user agentdeploy https://mcp.agent-deploy.ai/mcp
02 / Authorize AgentDeploy
/mcp
03 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client claude
04 / Verify the connection
claude mcp get agentdeploy
Official client documentation ↗
OpenCodeHosted OAuthAgent Skill
01 / Add the MCP server
opencode mcp add agentdeploy --global --url https://mcp.agent-deploy.ai/mcp
02 / Authorize AgentDeploy
opencode mcp auth agentdeploy
03 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client opencode
04 / Verify the connection
opencode mcp list
Official client documentation ↗
CursorHosted OAuthRemote MCP
01 / Add the MCP server

~/.cursor/mcp.json

{
  "mcpServers": {
    "agentdeploy": {
      "url": "https://mcp.agent-deploy.ai/mcp"
    }
  }
}
02 / Authorize AgentDeploy
cursor-agent mcp login agentdeploy
03 / Verify the connection
cursor-agent mcp list-tools agentdeploy
Official client documentation ↗
VS Code / GitHub CopilotHosted OAuthRemote MCP
01 / Add the MCP server

Command Palette → MCP: Add Server → HTTP → https://mcp.agent-deploy.ai/mcp → agentdeploy → Global

02 / Configuration alternative
{
  "servers": {
    "agentdeploy": {
      "type": "http",
      "url": "https://mcp.agent-deploy.ai/mcp"
    }
  }
}
03 / Authorize AgentDeploy

Chat → Configure Tools → agentdeploy → Start

Official client documentation ↗
Gemini CLIHosted OAuthRemote MCP
01 / Add the MCP server

~/.gemini/settings.json

{
  "mcpServers": {
    "agentdeploy": {
      "httpUrl": "https://mcp.agent-deploy.ai/mcp",
      "oauth": { "enabled": true }
    }
  }
}
02 / Authorize AgentDeploy
/mcp auth agentdeploy
03 / Verify the connection
gemini mcp list
Official client documentation ↗
Qoder CLIHosted OAuthAgent Skill
01 / Add the MCP server

~/.qoder/settings.json

{
  "mcpServers": {
    "agentdeploy": {
      "type": "http",
      "url": "https://mcp.agent-deploy.ai/mcp",
      "oauth": { "enabled": true }
    }
  }
}
02 / Authorize AgentDeploy
qoder mcp auth agentdeploy
03 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client qoder
04 / Verify the connection
qoder mcp list
Official client documentation ↗

Office and general agents

ChatGPTHosted OAuthRemote MCP
01 / Add the MCP server

ChatGPT web → Plugins → + → Add custom MCP server

02 / Configuration alternative
Name: AgentDeploy
Server URL: https://mcp.agent-deploy.ai/mcp
Authentication: OAuth
03 / Authorize AgentDeploy

Create as a plugin → review discovered tools → authorize AgentDeploy

04 / Verify the connection

Install the plugin → open a new Work chat → @AgentDeploy

Official client documentation ↗
WorkBuddyHosted OAuthMCP + Skill
01 / Add the MCP server

mcp.json

{
  "mcpServers": {
    "agentdeploy": {
      "type": "streamableHttp",
      "url": "https://mcp.agent-deploy.ai/mcp",
      "timeout": 30000
    }
  }
}
02 / Authorize AgentDeploy

Connectors → AgentDeploy → Connect → authorize in browser

03 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client bundle
# Include the installed brand Skill directory in this brand's connector.
# Keep auth_mode unset to use the standard MCP OAuth flow.
Official client documentation ↗
豆包工作 / Doubao WorkHosted OAuthRemote MCP
01 / Add the MCP server

In the Doubao Work desktop client (豆包工作), open Plugins · Skills · Partners → Plugins → + Add → New custom plugin. Name it AgentDeploy, select HTTP and OAuth authentication, and paste the server URL below.

https://mcp.agent-deploy.ai/mcp
02 / Authorize AgentDeploy

Save the plugin and connect. Sign in on the AgentDeploy page opened in your browser, select one or more organizations to authorize, and return to Doubao Work when authorization completes.

03 / Verify the connection

Save the plugin, start a new work task, and select AgentDeploy using @ or the Plugins menu. Send the prompt below to check access and inspect deployment candidates before choosing a node.

Use AgentDeploy to list the authorized organizations and deployment nodes that support Static. Show the candidates and ask me to confirm a node before deploying.
Official client documentation ↗
千问办公 / QwenWorkRemote MCPstdioAgent Skill
01 / Add the MCP server

扩展 / Extensions → 连接器 / Connectors → + 添加 / Add → 粘贴 JSON 配置 / Paste JSON

02 / Configuration alternative
{
  "mcpServers": {
    "agentdeploy": {
      "type": "streamable-http",
      "url": "https://mcp.agent-deploy.ai/mcp"
    }
  }
}
03 / Local stdio fallback
curl -fsSL https://install.agent-deploy.ai/mcp | sh
# Windows PowerShell: irm https://install.agent-deploy.ai/mcp.ps1 | iex

{
  "mcpServers": {
    "agentdeploy": {
      "type": "stdio",
      "command": "orbitmesh-mcp",
      "env": {
        "ORBITMESH_API_URL": "https://api.agent-deploy.ai",
        "ORBITMESH_INSTANCE_ID": "agentdeploy",
        "ORBITMESH_PRODUCT_NAME": "AgentDeploy",
        "ORBITMESH_TENANT_ID": "YOUR_ORGANIZATION_UUID",
        "ORBITMESH_API_TOKEN_FILE": "/absolute/private/token-file",
        "ORBITMESH_WORKSPACE": "/absolute/project/path"
      }
    }
  }
}
04 / Install the publishing Skill

Installation uses system tools, without Python or Node. Installed Skill names use the stable brand ID plus -publish, allowing multiple brands to coexist.

curl -fsSL https://install.agent-deploy.ai/skill/publish | sh -s -- --client qwenwork
Official client documentation ↗

One workflow across local and hosted agents

Both transports use the same immutable artifact, release, deployment, and readiness model.

01Discover

Select an authorized organization and an existing node.

02Build

Produce a static directory, choose a digest-pinned OCI image, or package a Worker module.

03Prepare

Upload or import an immutable content-addressed artifact.

04Deploy

Submit a generation-checked deployment and optional public exposure.

05Observe

Poll status, read bounded logs, stop, start, or roll back.

MCP, Skills, and authorization

Should I use Hosted MCP or stdio?

Use Hosted MCP for the simplest OAuth connection. Use stdio when the agent must package a local build directory or the client cannot complete hosted OAuth.

What does the Agent Skill do?

The Skill teaches the agent the AgentDeploy publishing workflow and tool order. It stores no credentials and does not replace MCP authorization.

Can one authorization cover multiple organizations?

Yes. Select one or more organizations during OAuth consent. AgentDeploy checks the chosen organization and its current permissions again on every request.

Can another MCP client use AgentDeploy?

Usually yes if it supports Streamable HTTP with OAuth 2.1 and PKCE, or local stdio. Use the same MCP endpoint and adapt the configuration shape required by that client.