Connect Didit to Claude With MCP

Full transcript of Connect Didit to Claude With MCP - Run KYC and Fraud Rules by Chat - Didit Academy video 4 of 10, 10:50. Watch on YouTube. Each paragraph links to the exact moment in the video.

​

What you’ll build

00:03 In this video, you’re going to connect Didit straight to Claude and then get real work done just by typing what you want in plain English. So, building a workflow, clearing your review queue, setting up fraud rules. All of it just from the chat. And the thing that makes this possible is the MCP, which lets an AI assistant securely use Didit’s own tools. So, let’s start with what this connection actually

​

The hosted MCP server

00:31 gives you and get it set up. So, Didit ships a hosted MCP server, so any MCP client can drive the whole platform. And so, the three products that you should already know by now map straight onto this one connector. So, there’s nothing really new to learn about Didit itself. You just type what you want. Claude calls the matching tool and shows you the result. So, anytime it’s about to change something, it’ll ask you to confirm if you do have permission set inside of Claude. So, one connector turns Claude into a full Didit operator. So, we could see here at the top, we’ve just got a a link here that we can add straight into Claude. So, I’m going

​

What the connector actually gives you

01:11 to click that in just a second. But, what do you actually get with this MCP? Cuz this isn’t just some wrapper around one endpoint. The connector exposes 108 different tools across 12 tool sets that covers sessions, workflows, KYB, fraud rules, cases, lists, webhooks. Basically, everything that you can do in this dashboard and all the data points, you can access through the MCP. And the cool thing is, you don’t actually need to put in any API keys. You can just do an OAuth authentication one click. I’m

​

Adding the connector in Claude

01:42 going to show you and walk you through that in just a moment. So, let’s add the Claude connector. So, we can add this in Claude desktop, right? We can add it in Claude code. We can set it up in CodeX. All sorts of different options for you there. All All so let’s just click this link and add the connector straight into Claude, shall we? So we can add a custom connector. This is in the web version. 02:02 We can add for now, and you can see it’s added the MCP, but we’re not connected to Didit yet. So we can click connect. You will sign in here. You’ll log in. You’ll authenticate. It’ll redirect you back to Claude. And so when you come

​

OAuth: authorizing without API keys

02:16 through to authorize the application, this is the screen that you’re going to see. And this is what permissions you’re giving it. All right, so I’m just going to hit authorize, and you can see at the top connected to Didit. So I’m just going to see what we’ve got here. So you can see if we click that plus button, you can see connectors. I can see Didit is connected

​

Listing the available tools

02:38 here. So if I say “What tools do I have access to?” We’ll be able to see the list of tools that we have access to here inside of Claude. So just like I mentioned earlier, we’ve got 121 tools that we’ve got access to in here. This is the list of everything. Now, don’t worry if this seems like a lot. This is just all of the the data points that we can tap into inside of Didit. So yeah, from here in this chat, you can just describe what you want, and it’s going to build it for you through the MCP.

​

Claude Code, Codex and ChatGPT

03:12 Now, if you are using a different tool like ChatGPT or Claude code, I mean Claude code is the same process. You’d have to just add the MCP URL, and then go through the authentication, signing with your account. It’s going to be very similar for ChatGPT. I’ve got a few prompts here that I’ve I’ve developed, so I’m going to throw them in, and we’ll see what we can create. Now, of course, you can create workflows. You can check your review

​

Querying your review queue

03:39 queue, all of this stuff. So So you can see we’ve got a permissions box show here. It’s going to ask us to use this specific tool. Going to click always allow, or you can change to just allow once. I’m going to hit always allow. So, we can see the result set is pretty large. So, perhaps we could have filtered it and said, “How many verification sessions do I have in review in the last 7 days?” Now, because I am using a demo account here, nobody is actually going in and approving any of the verifications that are in review. 04:11 So, the situation for you and your business, you’ll have somebody doing that daily, depending on the volume that you’re running. You’re not going to have a huge backlog. So, you can see it’s just pulling live data from our account, and let’s get into an actual build. So, I’m going to pop this prompt in here, “Build

​

Building a KYC workflow from one prompt

04:26 me a KYC verification workflow.” And then we’re going to name the checks that we want. So, I want an ID document check that reads the document, extracts the data, passive liveness, face match, AML screening. Once it’s built, set it as my default workflow. So, we do have another permission request that is going to ask you these permissions every time it’s using a new tool. Boom, so we’re done. 04:46 That workflow is built. That was super quick. It was probably like 10, 20 seconds. So, let’s go over to Didit and

​

Checking the result in the console

04:54 go into our workflows tab. Give it a refresh. And we can see KYC, liveness, AML, face match. And you can see last updated 1 minute ago. It’s created as a graph workflow.

​

Asking for a conversion-lift recommendation

05:12 Boom. Pretty cool. So, what else can we build here? Well, I’ve got another prompt that I think would be really useful if you’ve been using Didit for some time and you already have some data. “Take a look at my current workflows and my recent conversion data, then suggest one specific change that would lift my pass rate without weakening fraud protection. Please show me the exact update before you apply anything.” So, now again, I will say the account that I’m connected to is a demo account, so it doesn’t have a ton of actual data, but we’ll we’ll see what it can come back with. That took a few minutes, but we do have some results back. We could see what’s happening over the last 30 days. 05:52 443 sessions started, 84% conversion. But, look at the automated decision before human review. So, auto approved 338, auto routed to in review, so 24% of all sessions, and then 24 declined. So, we could see it’s just done an analysis of the approvals and the verifications. So, we could see the mechanism your default workflow ends in a blanket determine. So, I’m not going to read through everything here. It has suggested one change, replace the default workflow as blanket auto decide with the same AML score bands. So, low score AML false positives get auto approved instead of clogging review. 06:31 Kind of interesting. And you can run this analysis on your own account with your own data. And you can use AI to to help pull insights instead of having your team spend, you know, hours looking at your numbers and your reports. It’s great just to pull reports straight from Claude. You can pull specific numbers. You can even create a nice formatted PDF report straight here from Claude that you can download, that you can present in your meetings. So, this is just a really cool way of accessing all of your data and then using AI to analyze it all in one place. You can see it’s made a suggestion of what to change here. It’s validated the analysis, and then we can go ahead and ask it to implement these updates. Like you see, if the publish live step is what gave you pause, you can apply it as an unpublished draft. 07:22 Yeah.

​

Applying the change as an unpublished draft

07:26 So, we’re going to apply it as a draft. So, if you did want to do this with a live workflow, and you wanted to check everything first before you change something that’s live, then we can go ahead and do that, and you can see it’s updated. So, we’ll check that in a second. We’ll just set this other

​

Writing transaction monitoring rules

07:41 prompt off. to set up two real-time transaction monitoring rules. So a velocity rule that flags too many transactions in a short window and a structuring rule that captures amounts kept just under reporting threshold. Back test both against the last 90 days of my history and show me what they’d have flagged before we turn them on. So this can be super useful if you want to develop new rules that might make your verification process smoother, faster, and easier. 08:09 And then you can back test your data. So if we go over to didit, let’s just take a look at the workflow that we had created. We can see the version history and we can see boom, we’ve got a draft here in the version history. Got the V2, we can publish that and boom, it’s done that and you can see here at the bottom we do have the branching, which is the update that it was suggesting based off the AML screening. All right, so we’ve got a few questions here. Cord is asking

​

Answering the clarifying questions

08:35 us what the velocity rule should be. So how many transactions in what window? Let’s do three in an hour. What reporting threshold and just under band? So we’re going to flag Yeah, the 10K threshold. Let’s do all apps, all categories. Boom. So it’s just getting clarification on the specifics of the rules that we want to set up because I noticed that I didn’t actually specify that. 09:00 So it needs that information to create the rules and to run the back tests.

​

Back-testing against 90 days of history

09:05 So we can see it’s completed the back test. It did take a few minutes to go through all of this data. You know, looking back, I should have specified a specific time period or a specific amount of records to back test on. You know, if you have tens or hundreds of thousands of transactions inside of your account, then that might be something to consider or using a model like Sonnet that would be much quicker. So we can see it’s flagged a few users here where the transactions over three in one hour. And we’ve got the structuring multiple payments underneath that threshold. And you can see extrapolated to the full history, we’ve got a an analysis here. I’m not I’m not going to read through absolutely everything here, but you can see two things worth noting before you enable. 09:51 It’s given us some suggestions here. So, refine the rules, extend the back test. Uh we could even send a webhook to your Slack whenever these kinds of transactions come through or whenever somebody meets these specific rules. So, I’m just scratching the surface with what’s possible here with these AI tools when you integrate with Didit. You can use it for analysis, intelligence, for flagging specific issues, for crunching

​

Recap

10:15 data, for building, for editing, the whole works. And that’s the Didit MCP from end to end. One hosted connector, an OAuth login with no keys to paste, and 108 tools that let you build workflows, clear reviews, set up fraud rules, and figure out your declines, all just by asking. So, if you want to try it yourself, head over to didit.me, sign up for free, and add the connector just following the steps in this video.