MCP for Claude

@niravi/mcp is a Model Context Protocol server that exposes your Niravi corpus to Claude (and any MCP client) as tools — search, recall, chat, and every per-video read. The tools are generated from the same OpenAPI spec as the SDKs and routed through the SDK transport, so they track the API automatically.

It’s read-only by default (search/recall/analysis). No writes, no surprises.

Use with Claude Desktop

Add this to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "niravi": {
      "command": "npx",
      "args": ["-y", "@niravi/mcp"],
      "env": { "NIRAVI_API_KEY": "nv_..." }
    }
  }
}

Restart Claude Desktop. You’ll see the Niravi tools available — ask things like “search my videos for a sunset” or “what does video X say about pricing?” and Claude calls the corresponding tool.

Use with the Claude CLI / other MCP clients

NIRAVI_API_KEY=nv_... npx -y @niravi/mcp     # stdio server

Env var	Default	Purpose
`NIRAVI_API_KEY`	—	required — your `nv_…` key (scopes the corpus to its workspace)
`NIRAVI_URL`	`https://api.niravi.io`	override the API origin

What it exposes

The same surface as the API reference, minus the dashboard-only key-management routes: semantic + magic search, chat, per-video recall, and every read endpoint (transcript, scenes, faces, speakers, sounds, music, characters, screenplay). Each tool’s input schema comes straight from the OpenAPI spec.

Writes (upload / re-process / delete) are intentionally excluded from the published server. They can be enabled in a self-built copy — see the repo.

Source: github.com/Niravi-io/niravi-sdk/mcp · License Apache-2.0.