For developers

Your whole infrastructure — for your AI agents

Claude, Cursor or your own agent get 30 panel tools without a single SSH key. Dangerous actions run only with your approval, and every call is audited.

Try for free About the AI assistant

No SSH keys for AI Approvals for dangerous actions Every call audited

Claude · MCP: deploykin

> Deploy main to prod and clear the config cache

deploy_website site=demo.deploykin.com

✓ task #4812 — finished in 41s

run_artisan config:clear

✓ configuration cache cleared

run_command rm -rf storage/old-releases

⏸ awaiting your approval — a card in the panel and Telegram

The point

Access to the whole infrastructure — without the agent holding a single SSH key

The agent needs no keys, passwords or server credentials. It gets an MCP token — and the panel performs every operation itself over its own secured connection. Revoke the token and access closes instantly, with no keys left on servers to clean up.

A token instead of keys

One scoped MCP token — instead of handing out SSH keys and passwords to AI tools.

Secrets don't leak

.env files and configs reach the agent with secrets redacted, and SQL runs under separate managed database users.

Never more than you

A token can never do more than the person who issued it — the team role stays the ceiling.

Guardrails

Freedom for the agent, control for you

Three layers of protection between “the AI decided” and “it ran on the server”

Approvals

Sudo actions — arbitrary commands, database writes, firewall changes — always wait for your approval: a card in the panel or in Telegram. Arguments are frozen at request time — the model can't swap the command after you approve.

Denylist

Dangerous commands are blocked before execution: rm -rf /, mkfs, fork bombs, DROP DATABASE, UPDATE without WHERE. The critical protection tier can't be disabled even by the owner.

Scopes and audit

Read / operate / sudo levels on a specific server, site or database. Every call — allowed or denied — goes to the audit log, and denials trigger a notification.

Task manager

Every AI action is a task you can see

The agent does nothing “around” the panel: every action becomes a task in Deploykin's shared task manager — right next to your own deploys and operations.

Every task has a status, output and author — tagged “MCP” or “AI”
The agent waits for results itself: wait_for_task, get_task_output
Emergency stop: cancel_task kills the process right on the server
A failed chain can be restarted from the first failed step

app.deploykin.com / Tasks

AI Deploy demo.deploykin.com

Finished

MCP Dump of demo_prod database

Running

AI Restart php-fpm

Finished

AI rm -rf storage/old-releases

Awaiting approval

Quickstart

Three steps to your first “deploy it”

Create an MCP token

Issue a token in the panel and set scopes: read, operate or sudo — for a whole server, a single site or a database.

Add the endpoint to your client

"deploykin": {
"url": "https://app.deploykin.com/mcp",
"headers": { "Authorization": "Bearer dk_mcp_…" }
}

Just say it

“Deploy staging”, “why is prod returning 502?”, “dump the database” — the client calls the right tools itself.

Using claude.ai? Connect Deploykin via OAuth — no token copy-pasting.

What people automate via MCP

30 tools: everything you do by hand in the panel is available to the agent

Deploy from the editor

“Deploy this branch and wait for the result” — right from Cursor, without switching to the panel.

deploy_website · wait_for_task

Incident triage

“Why 502?” — the assistant reads the logs and metrics itself and suggests a fix.

tail_log · server_metrics

Database work

“Dump the database before the migration and show slow queries” — SQL without a console.

db_dump · db_query

Commands without SSH

artisan, WP-CLI and arbitrary commands — as the site's user, with an audit trail.

run_artisan · run_wp_cli

Pricing

MCP — on the Pro and Business plans

Reads — metrics, logs, statuses — don't consume the limit: only mutating actions count.

	MCP tokens	AI actions per month	Audit retention
Pro	2	200	30 days
Business	Unlimited	5 000	365 days

Works with your client

MCP is an open standard: one endpoint for everyone

Claude

claude.ai via OAuth, Claude Desktop and Claude Code

Cursor

MCP servers in the editor settings

Any MCP client

Anything that speaks Model Context Protocol

FAQ

Model Context Protocol is an open standard that lets AI clients connect to external services. Deploykin runs an MCP server: any compatible client gets 30 panel tools — from deployment to SQL.

Really. The agent only sees an MCP token. The panel performs all operations over its own secured server connection, SQL runs under separate managed database users, and .env files and configs are served with secrets redacted.

The command never reaches the server: the denylist blocks destructive patterns — rm -rf /, mkfs, fork bombs, DROP DATABASE, UPDATE without WHERE — before execution. The critical protection tier can't be turned off. And arbitrary commands only run after your approval in the first place.

Today programmatic access to Deploykin is MCP: an open protocol over HTTP that works without AI too. If you specifically need classic REST — write to us and we'll share our plans.

The tools, approvals and audit are the same. The assistant works right in the panel and in Telegram, while the MCP server exposes the same tools to external clients — Claude, Cursor and your own agents.