wealthmcp.ai

Comparison · AI Notetakers & Meeting Intelligence

AI notetaker MCP servers compared: Zocks, Otter.ai, Fellow, Fireflies and MeetGeek

Five servers in this directory connect an AI assistant to meeting recordings and transcripts. All five are official, all five are published by the notetaker itself, and all five read and write inside their own product rather than composing with anything else. What separates them is not capability so much as fit: one is built for advisory meetings, one can run on your own infrastructure, one also reaches into your email and CRM, and two are general-purpose meeting tools that happen to be useful.

There is a prior question worth settling before any of them. A transcript of a client conversation is client data, and depending on how your firm treats it, it may also be a record you are required to retain and supervise. Connecting an assistant to a notetaker means an agent can now read those conversations. That is a compliance decision before it is a tooling one.

At a glance

ServerStatusAuthHostingCovers
ZocksOfficialOAuthRemoteRead-only access to Zocks meeting intelligence (client/household context) for Claude and ChatGPT.
Otter.aiOfficialOAuthRemoteBidirectional meeting capture: acts as an MCP server for transcripts and history, and as an MCP client pulling from Gmail, Drive, Notion, and Salesforce.
FellowOfficialOAuthRemoteMeeting summaries, transcripts, action items, talking points, and calendar events queryable via Claude, ChatGPT, or Cursor.
Fireflies.aiOfficialOAuthRemoteTranscripts, meeting metadata, speaker info, and summaries queryable via Claude/ChatGPT (OAuth or JSON config).
MeetGeekOfficialAPI keyBothMeeting transcripts, summaries, action items, highlights, and insights via a hosted or self-hosted open-source MCP server.

Status, auth, hosting and coverage are reproduced from each server's directory listing, which reflects publisher documentation at the last review. Vendor claims are labelled as such and are not independently verified.

Where each one is the better answer

Zocks

Best for

The only wealth-native option here. Zocks is built for advisory meetings and returns client and household context rather than generic transcript text, and the connection is read-only by design.

Watch for

Narrow by construction. If your firm runs meetings outside client reviews, internal planning, vendor calls, recruiting, Zocks is not covering them. Check that its household model matches how your CRM actually organizes relationships before you rely on the mapping.

Listing and connection details →

Otter.ai

Best for

Breadth, and the only bidirectional entry on this page. Otter acts as an MCP server for its own transcripts and as an MCP client reaching Gmail, Drive, Notion and Salesforce, which makes it the closest thing here to a meeting layer that also sees context from elsewhere.

Watch for

That second direction is the thing to think hardest about. A notetaker that also reads your email, files and CRM has a far larger footprint than one that returns transcripts, and it is a separate authorization decision from letting an assistant read a meeting summary. Scope it deliberately rather than accepting the full grant.

Listing and connection details →

Fellow

Best for

Meeting management rather than transcription alone. Fellow returns agendas, action items, talking points and calendar events alongside summaries, which suits a practice running structured recurring reviews.

Watch for

Same general-purpose limitation as Fireflies. Its strength is meeting discipline, so the value drops sharply if your client meetings are not already run from structured agendas.

Listing and connection details →

Fireflies.ai

Best for

A general-purpose notetaker with a straightforward OAuth connection and speaker-level detail, useful when you need to know who said what rather than only what was decided.

Watch for

Built for meetings generally, not advisory meetings. Nothing in it understands households, accounts or suitability, so any mapping from a transcript to a client record is something your agent is inferring rather than reading.

Listing and connection details →

MeetGeek

Best for

The only option you can self-host. MeetGeek publishes an open-source server that runs hosted or on your own infrastructure, which matters if client conversations are not allowed to sit in a third-party cloud.

Watch for

API key rather than OAuth, so the key is a long-lived credential you store, rotate and revoke yourself. Self-hosting also moves the operational burden to you. Uptime, patching and access control become your problem rather than the vendor's.

Listing and connection details →

Endpoints and install commands are not repeated here, take them verbatim from each server's listing page, which is the version we keep current.

By advisor task

Pulling client meeting notes into a follow-up
Zocks, if having household and client context mapped for you is the point. Fellow or Fireflies if your meetings are not exclusively client reviews and you mainly need the transcript.
Client conversations cannot leave your own infrastructure
MeetGeek. It is the only one on this page you can self-host, and that is the whole reason to choose it.
Connecting meeting context to CRM and email
Otter.ai, because it also acts as an MCP client. Read its footprint note first. This is the largest grant on the page.
Knowing who said what in a joint client meeting
Fireflies, for speaker-level detail. Useful when a household has two decision-makers who do not agree.
Running structured recurring reviews
Fellow, whose agenda and action-item model is the closest fit to a repeatable review process.

Before you connect any of them

  • None of these five publish a public endpoint. All require account-level setup, so a firm administrator will be involved rather than an advisor wiring it up alone.
  • Only Zocks describes its access as read-only. For the other four, confirm what the connection can change rather than assuming it only reads.
  • Meeting content routinely includes people who never agreed to an AI assistant reading it. Recording consent and data-use disclosure are separate questions from MCP access, and connecting a server resolves neither.
  • Listings in this directory are informational. They are not endorsements, security reviews, or compliance advice.
All AI notetaker serversHow to connect