openkits as an MCP server

Every deterministic openkits tool is also exposed as a remote MCP server, so an AI agent (Claude, Cursor, Codex, …) can call it directly. One core, two shells: the web UI you already use, and this MCP endpoint.

Endpoint (Streamable HTTP):

https://mcp.openkits.dev/mcp

Tools

Add it to your client

Claude (claude.ai or Claude Desktop): open Settings → Connectors → add a custom connector and paste the endpoint URL:

https://mcp.openkits.dev/mcp

Cursor: add it to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "openkits": { "url": "https://mcp.openkits.dev/mcp" }
  }
}

Codex CLI: Codex talks to stdio servers, so bridge the remote endpoint with mcp-remote in ~/.codex/config.toml:

[mcp_servers.openkits]
command = "npx"
args = ["mcp-remote", "https://mcp.openkits.dev/mcp"]

Any stdio-only client: bridge with mcp-remote:

npx mcp-remote https://mcp.openkits.dev/mcp

Try it first with the MCP inspector, then connect to the endpoint over the Streamable HTTP transport:

npx @modelcontextprotocol/inspector

Privacy & limits

Public and anonymous — no account, no key. Per-IP rate limited, with a 1 MB input cap. Only deterministic, non-secret tools are exposed: there is no JWT signing/verification and no password generation over the network.

MCP client configuration changes quickly — if a step looks different, check that client's own MCP docs. The endpoint and tool names above are stable.