Superserve MCP
Superserve sandbox MCP server: an agent creates isolated Firecracker microVMs, runs commands in them and manages files
- 507 ★ on the repository
- Last commit
Broad access
Why this level
- The agent runs arbitrary commands and downloads files in a sandbox at your expense and with your key
- The API key manages every sandbox of the team and secrets can be bound to environment variables inside the machine
Install
- Works in
Manual install
claude mcp add superserve --env SUPERSERVE_API_KEY=ss_live_xxxxxxxxxxxxxxxx -- npx -y -p @superserve/mcp superserve-mcpLocal stdio mode; substitute your own key.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs arbitrary commands and downloads files in a sandbox at your expense and with your key
- The API key manages every sandbox of the team and secrets can be bound to environment variables inside the machine
This is third-party code. Review the repository files before installing.
What it does
Superserve is infrastructure for persistent and secure sandboxes for AI agents on Firecracker microVMs. The `@superserve/mcp` package opens it to any MCP client: Claude, Cursor, VS Code, Windsurf and Codex. The server runs locally over stdio through `npx` or remotely at `https://mcp.superserve.ai` with an API key in a Bearer header. There are twenty tools. A sandbox is created, updated, listed and deleted (`sandbox_create`, `sandbox_update`, `sandbox_list`, `sandbox_kill`), paused with its state preserved and resumed. Commands run in it with `sandbox_exec` returning stdout, stderr and the exit code, and files can be read, written, listed and downloaded as a directory archive. `sandbox_computer` drives a graphical desktop: screenshot, mouse, keyboard. There are also port publishing through a signed link, an outbound connection log, image templates and binding secrets to environment variables.
Who it is for. Agent developers who need a model to run risky commands, install packages and open a browser in a separate machine rather than on their laptop.
Good fit when
- An agent installs dependencies and runs a third-party repository's tests inside a sandbox rather than on your machine
- You need a graphical desktop with mouse and screenshots for tasks that cannot do without a browser
- The agent's result is published through a signed link to a sandbox port
- Outbound sandbox traffic is restricted by rules and checked in the network connection log
Not a fit when
- A one-off run of short code is enough: a local WebAssembly sandbox with no account or fee is simpler
- You need OAuth sign-in in Claude.ai or ChatGPT: the remote endpoint accepts a static key only for now
- Files larger than a few megabytes: reads are limited to 1 MiB and writes to about 4 MiB through the remote endpoint
Example request
Create a sandbox, clone my repository into it, run the tests, show the result and then delete the sandboxMCP
- Transport
- stdio, http
- Authentication
- API key
Environment variables
SUPERSERVE_API_KEYrequired, secret- Superserve API key starting with ss_live_, created in the console
SUPERSERVE_BASE_URL- Control-plane API base URL, https://api.superserve.ai by default
Limitations
You need an API key that starts with `ss_live_` and is created in the console at console.superserve.ai; clients do not inherit variables from the shell so the key goes into the config `env`. The remote endpoint does not support OAuth: Claude.ai, Claude Desktop custom connectors and ChatGPT developer mode cannot connect to it. File reads are limited to 1 MiB, writes through the remote endpoint to about 4 MiB and a directory archive to 10 MiB. Creating secrets through the server is intentionally left out of the server: values are entered in the console or SDK. The package version in the MCP registry is 0.3.4.
How to disable. Remove the superserve block from the client config (`claude mcp remove superserve`), delete the sandboxes and revoke the key in the console.
FAQ
How do I connect Superserve to Claude Code?
Locally: `claude mcp add superserve --env SUPERSERVE_API_KEY=ss_live_xxxxxxxxxxxxxxxx -- npx -y -p @superserve/mcp superserve-mcp`. Remotely: `claude mcp add --transport http superserve https://mcp.superserve.ai --header "Authorization: Bearer ss_live_..."`.
Where do I get an API key?
In the console at console.superserve.ai. The key starts with `ss_live_` and in MCP clients goes into the `env` block because clients do not take variables from your shell.
What happens to a sandbox after a pause?
Its state is preserved. `sandbox_exec` resumes a paused sandbox on its own, so `sandbox_resume` is usually unnecessary.
Can the agent create secrets?
No, creating secrets is intentionally not exposed in MCP so the value stays out of the agent transcript. `secret_list` lists stored secrets without values and `sandbox_attach_secret` binds them to an environment variable.
Is there an SDK for code?
Yes, `@superserve/sdk` for TypeScript and the `superserve` package on PyPI. The monorepo also holds a CLI, a console and documentation.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP bridge: a single run_python tool runs code in a container and calls your MCP servers
An MCP builder agent that picks servers from the official registry and composes them into your own agentic MCP server.
A Docker container bundling browser, terminal, files, VS Code and MCP services for running agent actions safely in isolation
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
An MCP server that runs sub-agents defined in Markdown through Codex, Claude Code, Cursor CLI and other CLIs from any MCP client
A read-only MCP server for the HeyClaude registry: an agent finds skills, MCP servers and hooks and gets ready install commands
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Superserve (this card) | yes | API key | 507 | |
| Ephemeral Sandbox | no | not required | 125 | a disposable code sandbox rather than a persistent microVM with a desktop and secret binding |
| Agent-Sandbox | no | key is optional | 232 | a self-hosted Kubernetes sandbox rather than a cloud service on Firecracker |
README in short
The main README is short: Firecracker sandboxes, the monorepo structure (console, CLI, TypeScript and Python SDKs, MCP), development commands and documentation links. Tool and installation details live in `packages/mcp/README.md`.