Beta

Any MCP server,
running in the cloud.

Install and run MCP servers on our infrastructure — no Node.js, Python, or Docker needed on your machine. Any npm or PyPI package works.

https://mcp.llmbase.ai/package/<package-name>/sse

Replace <package-name> with any npm package name. Pass API keys as query parameters.

github.com/LLMBaseAI/mcp-hosting

Quick start

1

Pick any npm MCP package

Any MCP server published to npm or PyPI works. No local install needed — we fetch and run it for you on-demand in an isolated container.

npm registry
# popular MCP packages on npm
firecrawl-mcp
@modelcontextprotocol/server-github
@modelcontextprotocol/server-slack
mcp-server-raycast
... any package that implements MCP
2

Add to Claude Code

One command in Claude Code, or a single JSON entry in Claude Desktop config. Pass API keys as query parameters — they're forwarded as environment variables.

Terminal
$ claude mcp add firecrawl \
    'https://mcp.llmbase.ai/package/firecrawl-mcp/sse?firecrawlApiKey=YOUR_KEY' \
    -t http
3

Done — it starts on demand

The first request spins up an isolated container with the package installed. Shared across your team. No runtime on any client machine.

Terminal
 Connected to mcp.llmbase.ai

  Server:   firecrawl-mcp
  Runtime:  isolated container
  Status:   ready

  No local Node.js or Python required.

Any package

npm & PyPI

Any MCP server published to npm or PyPI works out of the box — no configuration needed.

Isolated containers

Each server runs in its own Podman container with hardened defaults and resource limits.

Team sharing

One URL, usable by everyone on your team. No per-machine setup.

Open source

MIT-licensed. Self-host on your own infrastructure or use our public instance.

Open source · Beta · MIT licensed

Self-host or contribute

Clone the repo and run it on your own infrastructure, or contribute to the project on GitHub.

View on GitHub