For agentic companies

The VPS cloud that loves AI agents

CircuitNotion Cloud is agent-friendly from the ground up. An AI agent can create a Linux server, deploy to it, debug it and delete it over MCP or a REST API, with no SSH keys, consoles or YAML in the way. Build your agent product on servers your agents can drive themselves.

Start free Connect an agent

Why agents work well here

Agents are the main users of this platform, not an afterthought. Everything in the dashboard is in the API and the MCP server, and every response is shaped so that a model can act on it.

MCP-native

One URL gives Claude Code, the Claude API and other MCP clients 22 tools that cover a server's whole life.

Docs written for models

/llms.txt is a brief an agent reads in one call, and /api/v1/openapi.json describes every endpoint.

Errors that say what to do next

Every error is JSON with a stable code to branch on and a hint with the next step. Never an HTML error page.

Safe to retry

Send an Idempotency-Key and a retried create never makes a second server. ?wait=true returns once the instance has settled.

Strict by design

A misspelled or unknown JSON field is rejected with a clear message instead of being silently ignored.

A key for every agent

Limit a key to read, write, exec or files, lock it to one instance, or let it expire. Keys never mint keys or touch billing.

Sandboxed by default

Each instance is its own isolated container with hard CPU, memory and process limits, so code an agent writes stays in its box.

Every action on the record

The activity log names the key behind each create, command, deploy and delete, so you can see what each agent did.

What agentic companies run here

A sandbox per task or customer

Your product creates an instance per user, job or conversation over the API, labels it, runs code in it and deletes it when the work is done.

Coding agents that ship

A coding agent writes files, installs dependencies, runs the tests and starts the service, inside a Linux machine it fully controls.

Agent backends and APIs

Deploy the API behind your agent with a public HTTPS URL. A start command keeps it running and restarts it if it exits.

MCP servers and tools

Host your own MCP servers and tool APIs next to your agents. Your instances reach each other by name on a private network.

Workers and long jobs

Queue consumers, schedulers and crawlers on full Linux with root, a persistent /data disk, live logs and metrics.

GPU inference

Dedicated NVIDIA GPUs for the models your agents call, managed with the same API. Join the waitlist to tell us what you need.

One sandbox per task, in four calls

Give each task its own private machine, run the agent's code there, read the result and throw the machine away. The Idempotency-Key makes the create safe to retry.

REST · a disposable sandbox

API=https://cloud.circuitnotion.com/api/v1
AUTH="Authorization: Bearer $CNC_API_KEY"

# 1. a fresh, private machine for this task
curl -X POST "$API/instances?wait=true" -H "$AUTH" -H "Content-Type: application/json" \
  -H "Idempotency-Key: task-8841" \
  -d '{"name":"task-8841","template":"python-3.12","size":"small","public":false,"labels":{"customer":"acme"}}'

# 2. hand it the agent's code
curl -X PUT "$API/instances/task-8841/files?path=/app/solve.py" -H "$AUTH" --data-binary @solve.py

# 3. run it: exit code, stdout and stderr come back as JSON
curl -X POST "$API/instances/task-8841/exec" -H "$AUTH" -H "Content-Type: application/json" \
  -d '{"command":"python /app/solve.py","timeout_seconds":300}'

# 4. clean up
curl -X DELETE "$API/instances/task-8841" -H "$AUTH"

Made to be read by agents

/llms.txtThe whole platform in one plain-text brief: concepts, sizes and the calls to host an API.
/api/v1/openapi.jsonAn OpenAPI 3.1 description of every REST endpoint, for tool definitions and code generators.
/mcpA Streamable HTTP MCP server. Authenticate with a bearer API key and the tools are ready.

Questions from agent builders

What makes CircuitNotion Cloud agent-friendly?

It is designed for agents first. Everything you can do in the dashboard you can do over the REST API and the MCP server. The docs come in formats models read well, errors carry a code and a hint, creates can be retried safely, and every agent can have its own narrowly scoped key.

Which agents and frameworks work with it?

Any MCP client that supports remote Streamable HTTP servers, including Claude Code and the Claude API's MCP connector, and any framework that can make HTTP requests. Point it at the OpenAPI spec or /llms.txt.

Can we create one instance per customer or per task?

Yes, within your plan's limits. Create instances over the API, tag them with labels, list them by label, and give each one an API key locked to that instance.

Is it safe to let an agent run code it wrote?

Each instance runs in its own sandboxed container with dropped capabilities and hard CPU, memory and process limits, and only your own instances share your private network. Give agents keys with only the scopes they need.

Do you offer GPUs for agents?

Yes: instances with dedicated NVIDIA GPUs, managed with the same API and MCP tools. See GPU instances for what is online now, and join the waitlist to tell us which GPUs you need.

Give every agent its own server

Start on the free plan, connect your agent over MCP and let it deploy.

Start free with CircuitNotion