# Meeting transcript MCP: how to bring your calls into Claude and write from them (2026)

> Granola, Fireflies, Fathom, Otter, and Gong all have MCP servers for Claude. Here's what each one lets Claude see, and how to turn calls into content.

Source: https://contengi.com/blog/meeting-transcript-mcp
Author: Stefan Maritz
Published: 2026-11-08

Meeting tools now plug straight into Claude, and the trick is deciding which calls it should write from.

**TL;DR**

-   A meeting transcript MCP is a connector that lets Claude, Anthropic's AI model, search and read your recorded calls from inside a chat, so you stop exporting and pasting transcripts by hand.
-   Granola, Fireflies, Fathom, and Otter all run official MCP servers you add from Claude's connector settings, and Gong's server is built for sales insights and needs a Gong tech admin to switch it on.
-   Your plan decides what Claude can see: on Granola's free plan it reaches notes from the last 30 days, and the raw transcript tool is kept for paid plans.
-   Pointing Claude at every call you've ever had makes it a good searcher and a weak writer, because stand-ups and pricing calls sit in the same pile as the interview you want to publish from.
-   The setup that works for content keeps stable brand context in a knowledge base, keeps the transcripts you chose in their own section, and lets Claude pull both into one session, which is how Contengi runs it.

If you record your calls, you have a folder of them nobody opens. The good lines are in there, the customer who explained your product better than your homepage does, the throwaway opinion that would make a sharp LinkedIn post. Getting them out used to mean exporting a file, cleaning it up, and pasting a wall of text into a chat window while hoping the model found the right bit.

This post covers which meeting tools now have an MCP server, what each one lets Claude see, and the three-layer setup we use in Contengi to turn calls into on-brand content without drowning the model in every meeting you've ever had.

## What is a meeting transcript MCP?

A meeting transcript MCP is a server that exposes your recorded calls to an AI assistant through the [Model Context Protocol, the open standard for connecting AI apps to outside data and tools](https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro). Once it's connected, you ask Claude what a customer said about onboarding last Tuesday, and it looks the meeting up, reads the note, and answers in the same chat.

Your meeting tool still does the recording and transcription. The server gives Claude a door into the archive, and your login decides which rooms it can walk into. If the protocol itself is new to you, start with [what MCP is and how it connects Claude to your tools](https://contengi.com/blog/what-is-mcp).

## Which meeting tools have an official MCP server?

Granola, Fireflies, Fathom, Otter, and Gong all publish their own MCP servers as of November 2026. Community-built servers exist for most of them too, but the official ones use your normal sign-in and the vendor maintains them, so start there.

Tool

How you connect it to Claude

What to know

Granola

Claude's connector settings, or mcp.granola.ai/mcp in Claude Code

Free plan reaches the last 30 days, raw transcripts are paid-only

Fireflies

api.fireflies.ai/mcp, signed in with your Fireflies account

Transcripts, summaries, and action items, with an API key option for Claude Desktop

Fathom

Official connector in Claude, enabled by a team owner first

The same server at api.fathom.ai/mcp works in Claude Code

Otter

Listed in Claude's connectors directory

Finish the connection in Otter's web app

Gong

Set up by a Gong tech admin

Built around account and deal insights for sales teams

On Granola's free plan, Claude reads personal and public notes from the last 30 days, and the get\_meeting\_transcript tool is reserved for paid plans. Fireflies runs its server at api.fireflies.ai/mcp and signs you in with your Fireflies account. Fathom has an official Claude connector that a team owner has to enable before members see it. Otter sits in Claude's connectors directory, and for now you finish the connection in Otter's web app. Gong's server lets Claude ask questions about accounts and deals, generate summaries, and pull AI-generated insights, which makes it a sales tool first.

## What happens when Claude can read every call you've had?

It gets very good at finding things and noticeably worse at writing from them. A meeting MCP is scoped to everything your login can see, so your Monday stand-up, a pricing negotiation, a coaching session, and the customer interview you want to quote all sit in one pile. Claude has to search that pile every time, and the summaries it lands on flatten the speaker's phrasing, which was the whole point of using a transcript.

For content, the raw words are the asset. The way a customer describes their problem is the language your audience types into Google and ChatGPT. There's a permission question too, because a client call can be fine to learn from and wrong to quote.

I learnt the volume lesson the hard way. Users could upload freely into one knowledge base, one team added 400 PDFs, and the system got less reliable with every file. We removed the upload button and split what the agents read into layers.

## How Contengi brings transcripts into Claude

Contengi splits context into three layers and lets Claude pull from each one on demand.

**The knowledge base** holds what stays stable: positioning, audience, tone of voice, competitors, and hard rules. It's hand-curated and kept small on purpose, and [setting up a brand knowledge base for AI](https://contengi.com/blog/how-to-set-up-a-brand-knowledge-base-for-ai) is the first job we do with every customer.

**Transcripts** sit in their own section, apart from the brand files, and hold the interviews, podcast recordings, webinars, and voice notes you choose to keep. You drop a call into Contengi and it becomes source material with your phrasing intact, and Claude reads one only when the task calls for it.

**Live data** covers Search Console, analytics, and the web, fetched when the session runs because those numbers change daily.

Think of a restaurant kitchen. The meeting MCP is the walk-in fridge, everything's in there and some of it is labelled. Your Contengi transcripts are the prep station, where tonight's ingredients are already chopped and waiting. Nobody cooks out of the walk-in.

With the [Contengi connector added to Claude](https://contengi.com/features), a request like "turn last week's call about context layers into a LinkedIn post" runs in this order. Claude loads the brand context first, lists your recent transcripts, reads the one you named in full, drafts, runs the quality pass against your hard rules, and saves the draft to your library. This post was written that way. The call behind it was recorded in Fireflies, dropped into Contengi, and read by Claude in the same session as the brand rules.

### Where the meeting tool's MCP still earns its place

Keep it connected for retrieval. Ask Claude to find the call where a customer mentioned churn, or to pull every action item from a project. When a call turns out to be worth publishing from, drop it into Contengi so the next writing session starts from the good material. The same idea runs through [turning recordings into blog posts, social snippets, and show notes](https://contengi.com/blog/how-to-use-transcripts-for-content), and through the [YouTube transcript pipeline](https://contengi.com/blog/youtube-transcript-playbook) if video is where you do your talking.

## How to set it up this week

1.  **Connect the recorder you already pay for.** Add its official server from Claude's connector settings and sign in with the account that holds your calls.
2.  **Test retrieval before anything else.** Ask Claude for the action items from yesterday's meeting. If nothing comes back, check your plan limits and which workspace you're signed into, because Granola only reads your active workspace.
3.  **Pick five calls worth publishing from.** Customer interviews, podcast episodes, and talks you gave are good candidates. Leave internal syncs where they are.
4.  **Move those calls into Contengi.** Drop each one in as a transcript so it sits next to your knowledge base, ready to be named in a request.
5.  **Write from one call at a time.** Name the transcript in your request, let Claude load the brand rules before it reads anything, and run the draft through the [AI slop cleaner](https://contengi.com/ai-slop-cleaner) before it goes public.

## Frequently asked questions

### What is a meeting transcript MCP?

It's a server that lets an AI assistant like Claude search and read the calls stored in your meeting tool through the Model Context Protocol. You connect it once, sign in, and then ask about your meetings in plain language.

### Does Granola have an MCP server?

Yes. Granola runs an official hosted server at mcp.granola.ai/mcp that you add from Claude's connector settings. The free plan covers notes from the last 30 days, and full transcripts need a paid plan.

### Can Claude read my Fireflies transcripts?

Yes. Add api.fireflies.ai/mcp in Claude and sign in with your Fireflies account. Claude can then reach your transcripts, summaries, action items, and meeting insights.

### Is it safe to connect my meeting notes to Claude?

Each of the official servers sign you in as yourself and only return what your account can already open. Granola Enterprise admins can also switch off transcript access over MCP for the whole workspace. Treat client calls as internal material unless you have permission to quote them.

### Do I still need a content tool if my meeting recorder has an MCP?

The recorder gives Claude access to your calls. It doesn't give Claude your brand voice, your hard rules, or a place to keep the handful of calls worth writing from, and that's the job of a knowledge base and a hand-picked set of transcripts working together.