tl;dr: Gong MCP

  • Gong MCP returns AI-generated analysis of your calls, not the calls themselves. Gong’s documentation confirms transcripts are not returned.
  • The Gong MCP server exposes three read-only tools: ask_account, ask_deal and generate_brief. All exclude private calls.
  • Every Gong MCP request consumes Gong credits, and asking the same question twice consumes them twice.
  • Only a Gong tech admin can enable the Gong MCP server. Connecting Claude takes a manual custom connector setup on a paid plan, since Gong is not in Anthropic’s directory.
  • Gong MCP is worth switching on if you already pay for Gong. It is not a reason to buy Gong.
  • Tested by probing the public mcp.gong.io endpoint and reading Gong’s documentation on 1 September 2026, without a Gong account.
Оглавление

Gong MCP server allows your AI assistant to ask Gong detailed questions about all your sales deals, accounts, and pipeline. I bet if you’re a sales director, that sounds pretty good, right? Simply say, “Hey Claude, what deals are stalling?” Ding, you have an answer without clicking directly into Gong. Well, it’s not quite that simple, I’m afraid.

For transparency, Gong is in the same category as tl;dv, and is known as an incredibly sales-focused product. It’s designed and created for revenue teams and those who sell, or manage those who sell. It’s also a bit secretive, with no self-serve trial, no free tier; the only way to find out if it’s any good, and how much Gong costs, is to go the sales route. 

And there’s also been a bit of kerfluffle over their pricing model as of late, on 2 June 2026, Gong switched to metered AI pricing. This means that each seat on your account has a credit allowance that AI features use. So you pay for the tool, you pay for the seat, and then you have a balance of how much you can ask it. There’s been, understandably, a bit of grumbling over this. Worth pointing out that because Gong is so private about numbers (believe me, I’ve tried hard over the years!), a lot of this is third-party information. If you want the fuller picture, Gong pricing covers what buyers report paying.

This means that Gong’s MCP server lands squarely in the AI credit usage, with one of the tools racking up those credit spends. 

Again, full disclosure, I have no Gong account, and there is no way for me to test this firsthand without access. So I haven’t had access to poke around on the MCP server directly. 

I did try to get a bit smart, though, and on the 1st September 2026, I opened the terminal to probe the public endpoint at mcp.gong.io without having a login. Basically a server must publish how to connect to it before anybody can connect, so it’s public domain access I then had a look at Gong’s own documentation against its partner pages and against third-party listings of the same server.

What Is Gong MCP, and What Does It Do?

Gong MCP is a connector that allows an external AI assistant to query Gong and get answers. To use it, you simply connect it and work in Claude, ChatGPT or your AI of choice, and go through what Gong brings back on a deal or account, into your chat box.

The MCP is shorthand for Model Context Protocol, a shared way for AI assistants to communicate with other software without custom connectors for each pair. Gong runs its own server that facilitates that. Your assistant connects and it pulls the data.

There is one thing, however: unlike other MCPs like tl;dv that can pull the transcripts and calls when queried, Gong has taken a slightly different route.

The server returns AI-generated insights based on Gong data. It will take a look at and analyze calls and any communications within a specified time range and return a summarized response. So you’re getting the highlight reel effectively. 

It’s read-only and that means it can’t change anything within Gong itself which is pretty standard. 

 

Gong MCP documentation stating the server returns AI-generated insights and does not return call transcripts
Gong's help center confirms the MCP server returns summarized responses, not raw call data.

So if you see, Gong’s design choice is summarized responses based on Gong, rather than the actual raw data.

Gong also has something called the Gong MCP client, which is actually a separate product and works in reverse. This is Gong itself pulling in data from other products, using its AI features. For example, Gong says that AI briefer can reach into your support system so that it can flag up unresolved tickets prior to a renewal call. 

What Can You Actually Ask Gong MCP?

Gong MCP has three tools. Gong’s documentation calls them:

  • ask_account
  • ask_deal
  • generate_brief

 

The three Gong MCP server tools: ask_account, ask_deal and generate_brief
Gong MCP exposes three tools, each analyzing calls and emails within a time range you set.

ask_account will answer a single question about one account. Gong says that this can be used for asking about customer concerns and pain points, digging into risk, and also for account review preparation. This is the kind of tool that can be used before your QBR when you want to know the status before you head into that meeting.

ask_deal is simply querying specific deals. What’s stopping it from closing, where is it in the sales journey, and what’s been promised? This is the one for sales reps that they can ask before a call with a client, or when prepping their pipeline to share. 

generate_brief builds a summary of an account, deal or contact. This then generates an output, split into sections with stakeholders, themes, next steps, and whatever is needed to summarize the entire entity. It’s geared toward you if you’re a sales director and want a temperature check of any client or deal. 

The mechanics of how each tool is selected, as stated by Gong, are that the AI should not use ask_account or ask_deal for anything broad, and should not use generate_brief for anything that is specific. So it’s largely dependent on how the prompt in the AI is worded. 

Either option returns a filtered answer, put through Gong’s own AI, and does not bring in raw activity. There’s a parameter called includeSources, but this is switched off by default and would add links into the calls and emails. This means that unless you specifically ask for sources, it won’t switch on and you’ll be given an answer with no grounding. 

Of these tools, you’ll see there is nothing to lay out what calls have happened; there’s no tool to read transcripts, or even fetch them. 

What Gong MCP Will Not Give You

Gong MCP will not give you the exact words that someone has said. It is purely just an analysis of your calls. No raw data and no email body text.

Gong’s MCP gives you Gong’s own analysis of your call, and not the call itself.

The Gong REST API does offer this, but the MCP has been designed purely as a question-and-answer tool. 

So, hypothetically, you want to ask which accounts were negative about pricing last month. You’ll query the Gong MCP and will get back a paragraph naming accounts and giving more detail about the issues. You then ask for the verbatim transcript of this and it won’t be able to give it to you in the AI.

There is the includeSource option; if switched on, it will give you links to the source, but you cannot get that text into the assistant. It means that it cannot check itself.

Now, on a slightly different note, we have undertaken a fair amount of testing of tools that offer transcription and summarization. This includes languages such as Japanese and Brazilian Portuguese. What I have found is that sometimes the transcription isn’t 100% on some tools, and sometimes the summaries can be a bit off the mark. This means that, if for any reason, the transcription or summary Gong has produced isn’t correct, then that will get pulled through. No tool is totally accurate, but being able to pull that exact wording into the AI rather than relying on summaries is just a failsafe that needs to be taken into consideration. 

Gong documentation mapping MCP server tools to the APIs that consume Gong credits
Each Gong MCP tool runs on an API that draws down your Gong credit allowance.

Is Gong MCP Read-Only?

Yes. Gong clearly lays out that the MCP server only reads data; it cannot create, update, or delete anything in Gong or your CRM system.

One thing I do want to point out is that when I used the terminal to view the endpoint, the authorization metadata included a permission called mcp:write. Without having a full account to access it, I cannot state or know what this does. It could be a simple naming artifact rather than a hidden tool, but I can’t categorically confirm this either way.

How Much Does Gong MCP Cost to Run?

The million-dollar question, and one that is hard to answer fully. Each question that you ask the MCP server uses Gong credits. 

Gong shows the mapping on its documentation. The two ask_ queries run on the ask-entity API; generate_brief runs on the get-brief API. APIs consume Gong credits because they use AI to generate insights, and they state clearly in the documentation that you must check that you have sufficient credits before querying.

Gong documentation mapping MCP server tools to the APIs that consume Gong credits
Each Gong MCP tool runs on an API that draws down your Gong credit allowance.

What does that mean? Well, it means that every time you ask something, your credit goes down, and if you ask the same question twice, then you’re doubling your usage.

How much it uses depends on how much Gong has to read through. Gong names the time range and breadth as the things to pin down, and if you use a broad query with a large account, it’s going to eat up more of those credits.

This can mean… issues. Give ten reps access to this and tell them to ask it anything. The usage then depends on how much they want to know, and the more curious they are, the more it’s going to cost you. They actually publish a guide on how to reduce your credit consumption, so it’s clearly something that’s come up before.

Who Has To Set Up Gong MCP?

To get Gong MCP up and running, you’ll need the Gong tech admin. Tech admins are the people who enabled the MCP server for the tools, and they’ll need to be the ones to flip the switch.

So someone with admin rights creates the connection; they then select which AI tools it communicates with and how it behaves. Then you get your sales reps to connect. This can be done in stages so that you can make sure its fit for purpose.

The admin is also the one who needs to choose access. Gong offers personal access, so the rep will only be able to see and query what is in their own Gong dash, or a single connection means that anything in Gong can be recalled for anybody.

Gong documentation stating tech admins enable the Gong MCP server for supported AI tools
Only a Gong tech admin can switch the MCP server on, and only for the connectors Gong lists.

Now, in the screenshot above from Gong’s own documentation, you may notice something. There is no Claude. 

I also had a look on Claude’s connector directory (where tl;dv is), and wasn’t able to find it.

Gong states that you can use Claude along with ChatGPT and Copilot as a tool to pull in insights, but it is NOT a first-party connector.

So, to get it connected to Claude, you will need to do a manual setup. Your admin creates the connection with Gong, then the person adds it as a custom connector

Is Gong MCP Good for a Sales Team?

If you already have Gong, yes, switch it on. If you do not, Gong’s MCP is not the reason to choose Gong as your meeting recorder or your sales tool. 

If your goal is pipeline awareness across lots of accounts, then yes. Anything that requires you to be able to query the calls, dig into what is said and get more details, no. 

It seems counterintuitive to have designed it this way, but there are two reasons as to why. The first is volume. If you have a large-scale operation with multiple teams of multiple reps, that is a lot of calls and data. Gong’s logic is that summarizing them within the system itself means it can prevent omissions or cuts when they are transferred elsewhere.

The second is governance. Call recordings can include sensitive data like money or customer complaints. By Gong sending answers rather than the source text, it means that detail doesn’t leave Gong.

That design that keeps your call data locked inside Gong is also the same thing that prevents you from being able to spot-check Gong’s answer in your AI. 

If you’re a sales director tracking ten accounts, that’s fair, and you can still open Gong and check as you always have. But for a manager, fielding a team of many reps, coaching on things like objection handling, it makes it kind of tricky.

And if you’re not already a Gong customer, the cost of Gong itself is incredibly high before you even ask your first question. Gong doesn’t publish its pricing; it has no free trial, and each seat and platform costs money, onboarding costs money, and AI credits cost money. Each time the team asks a question, that costs money. 

 

Gong MCP vs tl;dv MCP

These two MCPs do very different things. Gong MCP is built for breadth. You can query your AI about an account and get a summary of what Gong’s read on it is. If you have a huge enterprise business, it’s a very quick way to get an answer fast.

tl;dv’s MCP server is built for this and retrieval as well. It can get the contents of the transcript when requested; its search can get into the body of the text rather than relying on a secondhand summary. You can ask what a buyer asked about an integration and get. the exact sentence in their own words.

There are three practical areas where you’ll notice a difference.

You can check the answer – Where tl;dv can retrieve the body text of calls, you can pull the exchange and read it. That means you can get actual detail and are able to coach on something specific.

Search works across the library – Remember the objection but not the customer? Then you can’t find it with Gong; you’ll need to know the name of the account to get that information. tl;dv can run a search across meetings, so ask it about “who mentioned this competitor last month” and it’ll be able to pull it.

Questions don’t cost extra – tl;dv doesn’t use AI credit each time you ask a question. That means if you ask the same question three times, then you’re not draining your allowance.

The difference is whether you can get back to what was actually said. Gong MCP tells you a customer raised a pricing objection. tl;dv actually shows you the sentence they said it in.

If you already run Gong, switch the connector on, and tell your team it returns Gong’s analysis rather than their calls, so nobody wastes an afternoon asking for quotes. If you are choosing a recorder now, decide on this: does your team need to know what is happening across accounts, or does it need to coach people on what was said? If the answer is yes to either, then tl;dv is your answer, and it has a much more transparent pricing structure

FAQs About Gong MCP

Yes. Gong runs a hosted MCP server at mcp.gong.io/mcp, separate from its REST API. It gives AI assistants three read-only tools that return insights about accounts and deals, and a Gong tech admin has to switch it on before anyone can connect.

No. Gong’s documentation states that raw data including call transcripts, message bodies and activity lists is not returned. You get AI-generated insights instead. Listings that describe a Gong MCP server returning transcripts are usually describing a different server built on Gong’s API.

Three. ask_account and ask_deal answer questions about one account or one deal. generate_brief produces a structured summary of an account, deal or contact and does not take a question at all.

Yes, indirectly. Every request consumes Gong credits, and repeat questions consume them again. generate_brief costs more than the other two because each section of a brief adds to the total.

Yes, but not in one click. Gong is not in Anthropic’s connectors directory and Claude is not on Gong’s connector list, so you add mcp.gong.io/mcp as a custom connector on a paid Claude plan, following Gong Collective’s setup guide.