What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude, Cursor, and Codex securely connect to outside data and tools. An MCP server exposes a set of tools the assistant can call — Loomscribe's server exposes six read-only tools over your Loom video library, so your assistant can look up transcripts, summaries, and action items instead of you copy-pasting them.

Which AI tools work with the Loomscribe MCP server?

Any MCP client that supports remote servers over streamable HTTP with custom headers — including Claude Code, Cursor, Codex CLI, and VS Code's MCP support. Tools that only speak stdio can connect through the mcp-remote bridge. Setup is the same everywhere: the Loomscribe endpoint URL plus your API key as a bearer token.

Is it read-only? What can a connected tool actually do?

Yes — strictly read-only. The six tools can list your notebooks and videos, read transcripts, fetch AI-generated summaries, action items, topics, and mind-map outlines, and run semantic search. There are no tools to modify, delete, or share anything, and every request is scoped to the account that owns the API key.

How is this different from cookie-based Loom MCP servers?

Community Loom MCP servers typically run locally and authenticate with a connect.sid browser cookie you extract by hand — which grants full account access and expires about every 30 days. Loomscribe's connector is hosted, uses revocable read-only API keys, and serves processed data those servers can't: cleaned speaker-labeled transcripts, AI artifacts, and embedding-based semantic search across your whole library.

Free accounts include 100 MCP tool calls to try the connector end to end — enough for several real agent sessions. Pro removes the cap entirely (fair-use rate limits apply) and allows unlimited API keys. Protocol setup like listing available tools never counts against the free allowance.

How do I revoke access?

Delete the key in Settings → Connect your AI tools and it stops working immediately. Keys are stored hashed, shown only once at creation, and you can see when each key was last used.

Do my videos or transcripts get shared with the AI tool's provider?

Only what your assistant explicitly requests through a tool call is returned to it — a transcript page, a summary, or search excerpts — over HTTPS, authenticated by your key. Loomscribe never pushes your library anywhere; the connector answers questions, it doesn't sync data.

✦ MCP connector — Model Context Protocol

Your Loom library, inside Claude, Cursor, and Codex

Last updated: July 23, 2026

Loomscribe ships a hosted MCP server for Loom videos. Connect Claude Code, Cursor, Codex CLI, or any MCP-capable assistant with one URL and an API key — and it can read your transcripts, pull summaries and action items, and semantically search everything your team ever recorded.

No local server to run. No connect.sid cookie to extract from your browser and re-extract every month. Keys are read-only, scoped to your account, and revocable in one click — so your assistant can look things up but never touch your recordings.

Ask your agent “what did we decide about the pricing page in last week's walkthrough?” and it answers from the actual recording — transcript, action items, and all — without you opening a single tab.

Start free See how to connect

100 free tool calls included · No credit card · Read-only by design

✦ The toolbox

Six tools your assistant gets, all read-only

Everything an agent needs to navigate, read, and search your video knowledge — and nothing that can change it.

list_notebooks

Discover your notebooks — titles, video counts, and summaries — so your assistant knows what's in your library before digging in.

list_videos

See every Loom inside a notebook with speaker, duration, and processing status. The map your agent navigates by.

get_video_transcript

Read the full cleaned transcript of any recording, paginated for long videos — verbatim quotes included.

get_notebook_artifacts

Pull the AI-generated summary, action items, topics, and mind-map outline for a whole notebook in one call.

get_video_artifacts

The same artifacts scoped to a single video — the fastest way to answer "what did this walkthrough decide?"

search_transcripts

Semantic search across every transcript you own. Ask in plain language; get the most relevant excerpts with their source video.

✦ How to

Connected in 3 steps

From Loom share link to an AI assistant that knows your recordings — a couple of minutes, one config snippet.

Step 1

Add your Loom videos to Loomscribe

Paste Loom share links into a notebook. Loomscribe extracts transcripts with speaker labels, generates summaries, action items, and topics, and indexes everything for semantic search.

Step 2

Create an MCP API key in Settings

Open Settings → Connect your AI tools and create a key. It's shown once — copy it. Keys are read-only, scoped to your account, and revocable anytime.

Step 3

Paste one config snippet into your tool

Add the Loomscribe MCP endpoint with your key as a bearer token in Claude Code, Cursor, or Codex CLI. Your assistant can now list notebooks, read transcripts, and search your videos.

Claude Code

claude mcp add --transport http loomscribe https://www.loomscribe.com/api/mcp/mcp --header "Authorization: Bearer YOUR_LSC_KEY"

Cursor — .cursor/mcp.json

{
  "mcpServers": {
    "loomscribe": {
      "url": "https://www.loomscribe.com/api/mcp/mcp",
      "headers": { "Authorization": "Bearer YOUR_LSC_KEY" }
    }
  }
}

Codex CLI — ~/.codex/config.toml

[mcp_servers.loomscribe]
url = "https://www.loomscribe.com/api/mcp/mcp"
http_headers = { "Authorization" = "Bearer YOUR_LSC_KEY" }

Replace YOUR_LSC_KEY with a key from Settings → Connect your AI tools. Tools that only support stdio servers can bridge via npx mcp-remote.

✦ Side by side

Hosted connector vs DIY cookie scrapers

Community Loom MCP servers prove people want this. Here's what a hosted, first-party connector changes.

Feature	DIY cookie-based server	Loomscribe MCP connector
Setup	Install a local server, extract your browser's connect.sid cookie, paste it into a config file.	Create an API key in Settings and paste one config snippet. Done.
Authentication	A borrowed browser session cookie that expires roughly every 30 days — then you extract it again.	Real API keys you can name, rotate, and revoke. Nothing to re-extract, ever.
Where it runs	On your machine, per machine — every laptop needs its own install and cookie.	Hosted by Loomscribe. The same URL and key work from any machine.
Transcripts	Raw captions, if the recording has them.	Cleaned, speaker-labeled transcripts — including AI speech-to-text for caption-less videos.
Search	Whatever the unofficial API exposes.	Semantic (embedding-based) search across your whole library, scoped to your account.
Safety	A session cookie grants everything your browser can do — including deletes.	Read-only by design. Six tools that can look things up and nothing else.

The DIY column describes the general pattern used by unofficial, cookie-authenticated Loom MCP projects. Loomscribe is an independent product and is not affiliated with Loom or with any community MCP server.

✦ Pricing

Try it free, go unlimited on Pro

The free allowance is enough to connect a tool and feel it work on your real recordings.

Free

100 MCP tool calls

Included with every account — several real agent sessions' worth. One API key; protocol setup and tool discovery never count against it.

Pro

Unlimited tool calls

No cap (fair-use rate limits apply), unlimited API keys, plus everything else in Pro — unlimited notebooks, videos, and chat.

See Pro pricing

✦ FAQ

Frequently Asked Questions

✦ Get started

Give your AI tools your team's video memory

Add your Looms, create a key, paste one snippet — and every walkthrough, standup, and demo becomes something your assistant can quote.

Start free See pricing