---
title: MnemoVi MCP Server
description: Connect MnemoVi to an AI agent over the Model Context Protocol so it can search and cite your own video archive.
canonical: https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base/mcp
locale: en
---

# MnemoVi MCP Server

> Markdown twin of https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base/mcp
> Fetch this instead of the HTML page: same content, a fraction of the bytes.
> Site structure and the full page list for agents: https://giglabo.com/llms.txt

Connect MnemoVi to an AI agent over the Model Context Protocol so it can search and cite your own video archive.

- Type: page

## What the local MCP server is

MCP (Model Context Protocol) is the open standard that lets AI agents plug into tools. [MnemoVi](https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base) ships an MCP server that runs beside the app, so your agent treats your video library like a database it can query.

- **It lives on your machine.** The server binds to localhost only (`127.0.0.1:5005`). The agent connects to it directly — no account, no upload, no relay in the middle.
- **Answers you can trust.** Every reply is grounded in real moments — the exact line that was spoken or the frame that was shown — each with a timecode you can open.
- **Works with the agent you already use.** Claude Desktop, Claude Code, Codex, Gemini CLI — anything that supports MCP tool connections.
- **You stay in control.** Search and read are on by default. Anything with a side-effect — adding videos, opening the app — is a switch you flip yourself, off by default.

## Connecting an agent

Copy the token from **Settings → MCP** in the app, then connect however your agent likes. That is the whole setup.

Claude Code — one command registers MnemoVi:

```
claude mcp add --transport http mnemovi \
  http://127.0.0.1:5005/mcp \
  --header "Authorization: Bearer <token>"
```

Claude Desktop — add the server to `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "mnemovi": {
      "command": "npx",
      "args": ["-y", "mcp-remote",
        "http://127.0.0.1:5005/mcp",
        "--header",
        "Authorization: Bearer <token>"]
    }
  }
}
```

Verify it is live by listing the tools:

```
curl -s http://127.0.0.1:5005/mcp \
  -H "Authorization: Bearer <token>" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
```

The server is bound to localhost and protected by a bearer token — a stray browser tab or another device on your network cannot reach it.

## What the agent can do

- **Search & discover** — find where something was said or shown, scope a search to a collection, or pull up "more moments like this one".
- **Read the content** — read the transcript and on-screen text around a hit, pull a whole talk to summarize, or grab a single frame to look at.
- **Map the library** — list every video with its source and length, and check whether a fresh import has finished indexing.
- **Add videos** (opt-in) — import a link or a file and track it until it is searchable. Off until you allow it.
- **Jump in the app** (opt-in) — open MnemoVi and seek the player straight to the moment found. Off until you allow it.

Ready-made workflows ship with it: **Find & cite** (answer a question using only your library, citing every claim), **Summarize a video** (pull the full transcript and return a clean summary) and **Analyze your niche** (survey the whole library and name its themes, weights and gaps).

In practice: you ask in plain words, the agent searches your library, reads around the hits for context, and answers with citations — a spoken quote at one timecode, an on-screen slide at another — each of which opens the player at that second in MnemoVi.

## Why it matters

- **Long-term memory for your agent.** Hundreds of talks, courses and clips become something the agent can recall on demand, not a folder it cannot open.
- **Private by construction.** Localhost-only, bearer-token auth, a strict origin guard. Your videos, transcripts and searches never touch a server — there is no server.
- **Grounded, never guessed.** The agent quotes real moments with timecodes. If it was not said or shown, it is not in the answer.
- **Yours, and it stays that way.** Pay once, works offline, and the index is an open format on your own disk.

## Where to go next

- [MnemoVi — Local Video Knowledge Base](https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base) — the app, its capabilities and pricing.
- [Ask your AI assistant — use case](https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base/use-cases/agent-mcp)
- [Mnemoria engine overview](https://giglabo.com/products/mnemoria)

## Related

- HTML version of this page: https://giglabo.com/products/mnemoria/applications/mnemovi-local-video-knowledge-base/mcp
- Site map for agents: https://giglabo.com/llms.txt