SimplaDocs
Free resource

Let Claude and ChatGPT Work Directly in Your Docs

If you're copying data back and forth between ChatGPT and your docs by hand, the MCP lets Claude or ChatGPT read, write, and build inside Coda and Superhuman Docs directly.

For Coda and Superhuman Docs users who want their AI tools to read and write documents directly, not just chat about them.

What you’ll be able to do

  • Ask Claude to build a full Coda document, tables and all, instead of dropping in text you have to reformat yourself.
  • Set per-tool permissions so write and delete actions need your approval before anything runs.
  • Connect outside context, like your calendar, so the AI can act on more than just the doc in front of it.

What’s inside

  • The full capability matrix: all 81 things the MCP can do inside Coda today, and what it still can't
  • Tips and tricks for using the MCP safely, including how to set tool permissions in Claude

Before you download

Does this work with ChatGPT, or just Claude?
Both. The video uses Claude, but the same MCP connection works in ChatGPT and any other AI tool that supports MCP.
Can the AI delete or change things without me approving it first?
Not unless you let it. Every tool, including write and delete actions, has its own setting: allow, ask first, or block.
What can it actually do inside Coda?
Create pages, tables, columns, and formulas, not just drop in text. The doc has the full list of 81 things it can do right now.
Full transcript

We're going to talk all about the MCP in Superhuman Docs, or Coda. I think it is one of the most transformative things I've seen in Coda in the last eight years. Maybe I say that a lot, but I really do believe it this time. Let's get to business.

So, what is an MCP, first off? If you're a developer, you're probably going to hate me for this definition, but I think the easiest way to understand it as a non-developer is: it simply allows AI models, Claude, ChatGPT, to interact with your software and systems. There's going to be a different MCP for each software, so there's an MCP for Superhuman Docs. In this video, I'm going to show you how the MCP works directly right here inside of Docs, really quickly and easily. You don't need any other tools or any other magic, you can do a whole bunch of stuff here. Then I'm going to show you how the MCP works inside of Claude, and how you can put your work on auto. And then I'm going to give you some best practices in terms of how to use it safely.

So let's look at this first. I've got a prompt here. If you want to use the MCP inside of Docs, just use your little purple stars down here in the bottom right-hand corner, and you can ask it to do anything. So let's ask it to search the web and write an article on this page that explains what an MCP is. Right now, at the time of recording, it does have access to the web, but not other tools yet. By other tools, I mean it can't access HubSpot, it can't access Salesforce, it can't access Slack. It can just access the doc itself and the web. So we're going to go and watch it work.

As it does work, note that the MCP can do all sorts of things. I've asked it, for example, to include an interactive button at the bottom of the article to indicate whether or not someone read the article, and to create a quiz with multiple questions. So it's not going to just dump text content, and that might be what you're used to as a user of AI models: text content. But the MCP is going to be able to do things like create tables, create columns, write formulas, almost anything that you can do inside of Coda, the MCP is going to be able to do as well.

It is preparing article content, so it's going to drop that in in just a second. But as you're waiting: if you're wondering what the MCP actually can do, beyond "almost everything," I do have a capability matrix, everything it can do. Here is an exhaustive list of things that it can and cannot do, 81 different things that it can do, and some things it cannot do, like automations, certain things with the UI, and certain things with table view. If you want this exhaustive list of things it can and cannot do, there's a link in the description below.

So let's see what it did for us. It went ahead, and it's written an article now that talks about what an MCP is. This is text content, as I mentioned, but it also dropped in a table. I can also ask it to do things like modify that table and add eight more questions, and you'll see it's thinking, and it's going to use different tools it has in order to add those different rows. It's taking quite a while, but I can expand here and see what it's actually working on, it lets me know what it's doing and why, like managing columns. But you'll see it did add those eight different things. So you can continue to chat with this Docs AI over here on the right-hand side to create your documents.

Now we're going to jump into Claude and see how the MCP works in Claude. Note that this would work in ChatGPT or other common ones as well, we just use Claude, so I'm going to show you that. I'm going to first show you one important thing about how this MCP works inside of these systems, that's really important for privacy and control. If I go down here to my settings, to my connectors, and Coda, you're going to see all the tools. The MCP is actually a bucket of tools, these are the tools or things that the AI is going to be allowed or able to do. I have three different options: I can allow it to run, I can say it needs approval to run, or I can block it entirely. So I can go in a granular way, and I can say, allow everything. Now, when I run a query inside of Claude, it's going to be able to use all these tools without explicit permission. Or you could say, actually, delete comments is a destructive action, these write and delete tools are more destructive, so I'm going to say you need approval, and then, of course, you could also block them entirely.

So it's worth it to spend a couple minutes inside this settings area to ensure that no destructive or crazy things happen. I'm usually more on the always-allow side, but definitely take things like delete pages and put them at need approval, so it doesn't accidentally delete a bunch of stuff or delete whole documents. So that is the MCP inside of Claude, and how you can actually tweak how it works.

So let's go watch it actually work in Claude as well. I'm going to go grab this prompt back here inside of this document. What I'm setting the stage for here is to show you why using it in Claude might be better than Docs AI. So, setting the stage: let's pretend that I'm a busy employee with a ton of calendar events this week, and I need to quickly know what's coming up, who I'm meeting with, and what I should focus on. I'm going to ask Claude for a look-ahead, a week preview for this week. It will give all my meetings at a glance. Based on the calendar invite information, name the top three meetings which seem to be of the highest priority, and create it as a Coda document using the MCP.

Notice that I'm telling it, and giving it some information, like: create tables configured as calendars for an at-a-glance view. So I'm using some Coda-specific vocabulary, so it doesn't just go wild and create something weird, I want it to create something curated. I also told it that it can use Google Calendar, web search, and Coda. So one thing I am using is Google Calendar. If I go ahead and use Docs AI, it does not have access to my calendar, it doesn't have access to other systems. So when I'm using this inside of Claude, it's going to be able to find and use those things.

So here we go, and I'm going to now paste it and see what happens. You'll notice that it's searching available tools, it's trying to find out what it should do. It's listing calendar events and invites, it's pulling the list of users and calendars. So it's doing a lot of pieces, and this does take a while and can take quite some time, because I've asked it to do a lot. I've actually done this exact same prompt just a little bit before, and I have a doc for it. This doc has not yet been touched by me, this was completely created by the MCP. So notice here, the MCP could do far more than just create a table or a page, it can create whole documents and whole systems. Here is what it was able to create: a week-at-a-glance. Let's go back and actually find those rows that it created. You'll see here it created rows based off of my calendar, it found a discovery call with Ramp. It also provided background on the individuals that I'm going to meet with, here's key people that I'm meeting with. Note that this is all fake, these were all pretend events that I created on my calendar.

So what it did is it did some research, and it let me know who each of these individuals are: hey, I'm meeting with this guy, this is who he is. It gave me a whole calendar as well, with sourcing and caveats, and the top three priority meetings. So this is why using the MCP, especially in Claude or ChatGPT outside of Docs, can be really, really powerful, because it can use and connect to other tools. And then, if you're using it correctly, you can also run these on scheduled tasks. For example, I could create different scheduled tasks that actually create these weekly write-ups every single Monday.

So that is your intro to the MCP. It is just an intro, there is so much that the MCP can do, and so much that we at SimplaDocs are using it for internally. If you want to know more about that, go to the description below, you can get access to a PDF and some resources about some tips and tricks on using the MCP.