Keep working in Claude Code, Codex, Cursor, or the coding agent you already use. The Anyhost skill gives it a safe way to configure the runtime, deploy your existing remote MCP server, check health, and return the URL.
For HTTP-based remote MCP servers · GitHub repository required · production changes require confirmation
The boundary is deliberate. Your coding agent understands the application; Anyhost gives that agent narrowly scoped operations instead of raw provider credentials.
Start with the runtime. Add only the services the server actually needs; generated variables are injected into the target environment after provisioning and deployment.
A deployment job pinned to a pushed commit, with structured terminal state and a public HTTPS route.
Separate dev and prod values. Secret contents stay masked; production mutations require explicit confirmation.
Your agent can read deployment status, runtime logs, public URL and health evidence through the CLI.
Create project-scoped resources and receive generated DATABASE_URL or REDIS_URL at runtime.
S3-compatible storage with a project-and-environment scope and generated bucket, prefix and region values.
Bind a custom hostname, validate DNS and certificate state, then verify the public route after cutover.
The job is complete only when the exact deployment reaches a terminal state and the supported smoke path passes. The result remains machine-readable for the next debugging conversation.
Your coding agent fetches the official setup prompt and installs the project-local Anyhost skill package.
The agent confirms the transport, start command, health path, required secrets and managed resources.
Anyhost builds the pushed commit, starts the runtime and returns structured status, URLs and logs.
Only after dev verification does the agent ask for explicit approval to create or change production state.
The prompt asks the agent to prove the repository's current MCP transport before it promises a remote endpoint.
Fetch https://anyhost.app/agent-setup/prompt.md and follow it to set up Anyhost in this repository. Then prepare and deploy this existing project as a remote MCP server on the Anyhost dev environment. Inspect the repository first and tell me which MCP transport it already supports. Do not claim remote hosting if it only supports stdio; make any required application change explicit. Configure only the required environment variables and managed services, deploy the pushed commit to dev, verify the supported health path and one real MCP request, and show me the deployment receipt and logs evidence. Ask for my explicit confirmation before creating or changing anything in production.
No. Anyhost hosts an application that exposes a supported HTTP endpoint. If the repository only implements stdio, your coding agent must add a remote transport and verify it before deployment.
No. Tool definitions and application behavior remain in your repository. Your coding agent may inspect or edit them because you asked it to work in the repository; the Anyhost control plane receives deployment and resource operations.
Yes. Your agent can request managed resources through the Anyhost CLI. Ready resources inject generated environment variables into new deployments, without copying provider credentials into the repository.
Authentication belongs to the application. Anyhost can store its required secrets and run the service, but it does not invent an authentication policy or make an unauthenticated MCP server safe.
No. Production deployment and production resource changes require explicit confirmation at the CLI boundary. The agent can inspect production state without changing it.
Start with dev, verify one real MCP request, and only move to production when the evidence is good.