Model Context Protocol

Noverva as an operating layer for your AI client

Your sites, their health, audit findings and content opportunities, and crawl requests are available to an MCP-capable client — under your permissions, with confirmation for anything consequential.

Start your 14-day trial Read the docs

Authorisation

MCP does not get its own powers

The most common way an integration layer becomes a security incident is by running as a service account with broad rights. Noverva's MCP server has no ambient authority at all.

It is you, not a robot

Every call resolves to a real principal with real roles. Queries run through the same row-level security as the browser session. A site you cannot see does not appear.

Writes ask first

Anything that changes state or fetches from your site requires explicit confirmation. Today that is one tool, request_crawl, and it does nothing until the call confirms.

No back door

A tool you could not run in the interface fails in MCP with the same error and the same reason. There is no path through MCP that is not also a path through the product.

What is exposed

Resources, prompts and tools

The specification distinguishes three things a server can offer. Noverva offers all three, and this is the complete list of each.

Resources

Read-only context a client can attach to a conversation: the list of your sites, and one site’s audit. Addressed by URI.

noverva://sites/{siteId}/audit

Prompts

Parameterised workflows you invoke deliberately: summarise a site’s audit (audit_summary), review its opportunities (opportunity_review), or get a short briefing on it (site_briefing).

audit_summary(siteId)

Tools

Actions with typed inputs and outputs. Read tools are available according to permission; the write tool is gated behind confirmation.

request_crawl(siteId, confirm)

Tool reference

Read freely. Write deliberately.

Read tools follow your permissions with no extra ceremony. Write tools state their consequence before they act.

Read

list_sites: Sites in your workspace that you can see. An empty list may mean there are none, or none visible to you.
get_site_health: Crawl status and open findings by severity for one site, from its most recent completed crawl.
list_findings: SEO findings for a site, each with the rule that produced it, its evidence and what to do.
list_opportunities: Scored content opportunities for a site, with the breakdown behind each score. Scores do not include search volume or difficulty.

WriteConfirmation required

request_crawl: Queues a fresh crawl of a site. It fetches the site's pages, so it requires explicit confirmation.

Connecting

Point your client at Noverva

The endpoint is below. It currently authenticates with your signed-in Noverva session, and MCP access is part of the plans that include the API.

Endpoint

POST https://noverva.com/api/mcp

Every call runs as you, through the same permissions as the product. Access with an API key from a standalone MCP client is planned.