Hostacker MCP
Coming soonCloud infrastructure, directly from your AI agent
Hostacker MCP will let developers discover, quote, deploy and manage infrastructure directly from compatible AI coding agents.
User
“Deploy everything this project needs for production.”
Agent
Hostacker recommendation
- Compute$11.99
- PostgreSQL$9.99
- Redis$4.99
How it will work
Connect once
Add Hostacker to your AI coding environment, approve the scopes you are comfortable with, and your agent gains a safe, priced view of your infrastructure.
Then ask:
- “Create a server for this project.”
- “Deploy PostgreSQL.”
- “Find the cheapest GPU for this workload.”
- “Show my infrastructure costs.”
- “Restart production-api.”
Cursor
One-click install from the MCP directory
Claude Code
Add via CLI or project config
VS Code
Copilot Chat MCP support
MCP-compatible clients
Any client speaking the protocol
Request flow
Cursor / Claude Code
Your AI coding environment
Hostacker MCP
Tools, quotes and approvals
Hostacker Control Plane
Policy, budgets and audit
Compute / AI / GPU / Apps
Provisioned resources
Your agent is never given a path to an infrastructure provider. Every request is designed to pass through the Hostacker Control Plane, where policy, budgets and audit logging are enforced.
Hostacker MCP will never expose provider credentials directly to the agent.
Safety
Safe by design
Giving an agent the ability to spend money and change production is only reasonable if the boundaries are real. These are the controls Hostacker MCP is being built around.
Scoped permissions
Grant read, restart or resize independently. Destructive actions start disabled and stay disabled unless you turn them on.
Spending controls
Hard monthly ceilings and per-action thresholds, so an agent cannot spend past a limit you did not set.
Approval thresholds
Anything above your threshold pauses and waits for a human to approve it in the dashboard.
Audit logs
Every tool call is designed to be recorded with the prompt, the resource, the cost and the approving identity.
Resource-level permissions
Restrict an agent to a single project, environment or tag so staging work cannot touch production.
No provider credentials
Hostacker MCP will never expose provider credentials directly to the agent.
All values shown in this section are demo policy, not an active configuration.
Permissions you can read at a glance
Each capability is granted independently. Read-only access is the default, destructive operations start disabled, and anything that changes cost or capacity can be forced through a human approval step.
- Scopes are per key, per project and per environment.
- Approvals arrive in the dashboard with the full agent context.
- Every tool call is written to an immutable audit log.
- Hostacker MCP will never expose provider credentials directly to the agent.
Example agent policy
Demo- Monthly budget
- $100.00
- Auto approve below
- $5.00
- Require approval above
- $10.00
- Server deletion
- Disabled
- Billing purchases
- Approval required
Servers
- ReadAllowed
- RestartAllowed
- ResizeApproval required
- DeleteDisabled
Billing
- ReadAllowed
- PurchaseApproval required
Tool surface
A small, explicit set of tools
Hostacker MCP is not a shell. It exposes a bounded set of infrastructure operations, each with a schema, a price estimate and a policy check.
list_resourcesInventory of servers, apps, endpoints and GPU jobs.
quote_infrastructurePriced plan for a described workload.
deploy_serverProvision compute under the active spending policy.
deploy_appInstall a catalog application on new or existing compute.
restart_resourceRestart a resource the agent is allowed to touch.
get_costsMonth-to-date spend and forecast by category.
Planned endpoint
https://mcp.hostacker.comBe first when Hostacker MCP ships
Create an account now and we will enable MCP on it as soon as it is ready.