MCP Server
Use deplo.sh from Claude, Cursor, OpenClaw, and any MCP-compatible agent host.
The Deplo MCP server exposes deplo.sh as a set of MCP (Model Context Protocol) tools. Any MCP-compatible agent host can use it — no installation required. It is hosted at mcp.deplo.sh.
Why use the MCP server?
If you use Claude Desktop, Cursor, Windsurf, OpenClaw, or any MCP-compatible AI host, the MCP server is the zero-friction way to give your AI assistant deployment powers:
- No code to write — your AI calls
deployas a tool, just like it calls "search" or "read file" - Conversational deployment — say "deploy this report" in natural language
- Full project management — list projects, rollback deployments, delete projects — all from chat
- One URL, zero install — add
https://mcp.deplo.shto your config and you're done
When to use MCP vs. other methods:
| Use MCP when... | Use something else when... |
|---|---|
| You're chatting with Claude, Cursor, or another MCP host | You're building a custom agent → use the SDK or REST API |
| You want the AI to decide when and what to deploy | You need precise control over deploy params → use the CLI |
| You want zero setup — just add a URL | You're integrating into a CI/CD pipeline → use the CLI |
The difference: with vs. without deplo.sh
Without deplo.sh: "Build me a landing page" → AI writes HTML → you copy it → create a repo → push to GitHub → configure Vercel → wait for build → get URL. (5+ minutes, manual steps)
With deplo.sh MCP: "Build me a landing page and deploy it" → AI writes HTML → AI calls deploy → you get a live URL. (30 seconds, zero manual steps)
Configuration
Claude Desktop — add to ~/.config/claude/claude_desktop_config.json:
{
"mcpServers": {
"deplo": {
"url": "https://mcp.deplo.sh",
"env": {
"DEPLO_KEY": "sk_live_xxxxxxxxxxxxxxxx"
}
}
}
}Cursor — add to your Cursor MCP settings:
{
"deplo": {
"url": "https://mcp.deplo.sh",
"env": {
"DEPLO_KEY": "sk_live_xxxxxxxxxxxxxxxx"
}
}
}OpenClaw:
url: https://mcp.deplo.sh
env:
DEPLO_KEY: sk_live_...Available tools
Deployments
| Tool | Description |
|---|---|
deployments_create | Deploy files and get a public URL. Auto-creates the project if the slug is new. |
deployments_list | List all deployment versions for a project, newest-first. |
deployments_get | Get full details of a specific deployment version. |
deployments_promote | Make an older deployment the live version (rollback). |
deployments_archive | Archive an old deployment to free up storage. Files are permanently deleted. |
Projects
| Tool | Description |
|---|---|
projects_list | List all projects in the workspace. |
projects_get | Get full details of a project by ID or slug. |
projects_create | Create a new project with optional gate and TTL settings. |
projects_update | Update project settings: name, domain, gate type, password, or default TTL. |
projects_delete | Permanently delete a project and all its deployments. |
Gate & Leads
| Tool | Description |
|---|---|
gate_list_leads | List captured email leads for a project using the email gate. Supports pagination. |
Domains
| Tool | Description |
|---|---|
domain_verify | Check DNS propagation for a project's custom domain. |
Tokens & Account
| Tool | Description |
|---|---|
tokens_list | List all API tokens in the workspace. |
tokens_create | Create a new scoped API token (shown only once). |
tokens_revoke | Permanently revoke a token by ID. |
account_me | Return info about the authenticated key: workspace, scopes, project restriction. |
Example prompts
Deploy a report:
"Generate a summary of our Q1 sales data, format it as an HTML report, and deploy it to deplo.sh under the project
q1-sales. Return me the URL."
The agent calls deployments_create with type=static_file, one index.html, and returns the live URL.
Build and deploy a dashboard:
"Create an interactive data visualization of the attached CSV file, build it as a static site, and deploy it to deplo.sh."
The agent generates the HTML/CSS/JS, calls deployments_create with type=static_site, and returns a browsable URL.
Rollback:
"The latest deployment of
weekly-reportbroke something — roll back to the previous version."
The agent calls deployments_list to find the previous version, then deployments_promote to restore it.
Set up a lead-capture gate:
"Make the
marketing-landingproject require visitors to enter their email before they can see the content."
The agent calls projects_update with gate_type=email. Visitors will see a gate page and must submit their email to proceed.
Export leads:
"List all the email leads captured from the
marketing-landingproject."
The agent calls gate_list_leads with the project ID, returning each lead's email and capture timestamp.
Tip: Create a deploy-only token for MCP integrations using
deplo tokens create "MCP Agent" --scopes deploy. This limits what the agent can do if a prompt injection occurs.
How it works
The MCP server is hosted at mcp.deplo.sh and communicates with your agent host over HTTP. It forwards tool calls to the deplo.sh REST API using the DEPLO_KEY you provide. No local installation is required.
What's next?
- Quickstart — deploy your first file via the CLI
- Deploy with Cursor — step-by-step MCP tutorial for Cursor
- Deploy from Claude Desktop — step-by-step MCP tutorial for Claude
- REST API — the HTTP API the MCP server calls under the hood
- What is deplo.sh? — platform overview and design principles
- Authentication — creating and managing API tokens