Model Context ProtocolStreamable HTTP

Isvara, in your agent.

The grounded advisor and the operations score are also a Model Context Protocol server. Add one URL and your own agent, Claude or any MCP client, can ask iO, score a measured signal, and read the field guide, all grounded and cited. It runs on the same backend as the site, so the answers, the score and the citations are identical.

Connect

https://isvara.io/api/mcp

Add it as a connector in Claude, or run it locally over stdio:

npx mcp-remote https://isvara.io/api/mcp

Tools

Two ways to act.

ask_isvara(question, chapter?)

The grounded operations advisor (iO). Answers about running a VMware Cloud Foundation or vSphere private cloud, grounded in the Broadcom field guide, citing the exact sections as /read links. It may also consult the current Broadcom tech docs when the guide lags.

{ "question": "What is a healthy CPU ready percentage, and why does it matter?" }

ops_score(metric, value)

The signature four-band 0-100 operations score for one measured signal. metric is cpu_ready (%), mem_latency (%), disk_latency (ms) or dropped_packets (count); value is the measurement. Pure maths, the same curve the site and iO use.

{ "metric": "disk_latency", "value": 40 }  ->  0/100, red band

Resources

The guide, readable.

isvara://indexEvery cite-able chapter across the three books, each with its /read deep-link.

isvara://thresholdsThe service-level indicators, their thresholds, and how the 0-100 score is derived.

isvara://book/{id}The source Markdown of a book. Ids: exec-summary, part-1, part-2, part-3, part-4.

Adding it

One URL.

Remote (recommended). In Claude, add a custom connector with the URL https://isvara.io/api/mcp. No install.

Local (stdio). Bridge to it from a desktop config with mcp-remote:

{
  "mcpServers": {
    "isvara": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "function(){throw Error("Attempted to call MCP_ENDPOINT() from the server but MCP_ENDPOINT is on the client. It's not possible to invoke a client function from the server, it can only be rendered as a Component or passed to props of a Client Component.")}"]
    }
  }
}

Grounded and scoped. ask_isvara reuses the site’s system prompt, model routing and injection guard, so it is the same public, rate-limited surface as the chat on this site. The server is deliberately scoped to the advisor, the operations score and the guide; the other Tools-page instruments stay on the site.