How to Use SendPulse’s MCP with Your Chatbots and CRM System

We’re excited to introduce our official Model Context Protocol server. This game-changing enhancement lets you connect an AI tool to your SendPulse account. Then it instantly becomes your very own personal AI agent that gets your chatbot- and CRM-related tasks done in seconds.

In this post, we’ll explore how to make your AI tool go beyond its chat interface.

What is an MCP server, and how does it work?

AI models can suggest great ideas, but they can’t bring them to life on their own. With no real-time data or ability to take actions, like update deals or send messages, their potential stays limited.

The Model Context Protocol (MCP) changes that. It is a universal, open standard that bridges isolated AI systems with external tools and real-time data. It helps your AI-powered tool become an independent AI agent that turns static recommendations into dynamic, nuanced actions and handles everyday tasks for you.

Once you’ve connected the MCP server, you can use the tools it offers in chats. Here’s how it operates:

1. You ask your AI assistant, such as OpenAI, Claude, or Cursor IDE, something as usual.
2. It forwards your request to SendPulse’s MCP server.
3. The MCP server calls the API to fetch data or perform an action.
4. Your AI assistant sends the result back to your chat.

Think of it as an interpreter between the human language your AI assistant speaks and the technical language the API uses. You don’t need to get into any under-the-hood details – you only need a clear idea of what you want and what goes into it.

Currently, SendPulse’s MCP server integrates with our chatbot builder and CRM system. Integrations with our email marketing service and pop-up builder are just around the corner. Soon, you’ll be able to access self-running AI agents capable of managing complex workflows, including omnichannel campaigns.

Through MCP, AI agents can only take specific actions that are listed in your toolset. You’ll see the full list of actions when connecting our MCP server.

In our chatbot builder, AI agents can:

• access account and chatbot information;
• view statistics and tag lists;
• send campaigns;
• review and launch flows;
• view and add notes;
• manage contacts;
• browse chats and messages.

To see the description of each tool and required request parameters, click the arrow next to it.

Our MCP server is designed to help you automate routine workflows in your SendPulse account, stay efficient, and reach your goals sooner.

5 ways to use MCP with chatbots

When interacting with AI tools, the key is to start with a clear, specific prompt. If the output doesn’t meet your expectations, you can always refine and rerun it. Just make sure the intended command is actually available — for instance, your AI assistant won’t be able to create a chatbot flow, no matter how explicit your prompt is, if there’s no corresponding tool.

To give you an idea of what AI can do with our MCP server, we’ll use an Instagram chatbot of a fashion and apparel brand.

Analyze conversations and discover insights

Unlike traditional metrics, such as CTR, open rates, or conversions, chat conversations reveal the “why” behind user actions, for instance, curiosity, doubts, fears, or the need for more information. That’s why analyzing chats can help you get deeper insights:

• which topics or phrases elicit an emotional response;
• where users drop off from the flow;
• what information helps them make purchase decisions;
• how behavior differs across audience segments.

Thanks to the MCP server, AI can securely access your chatbot conversations, analyze them, and share valuable insights to help you optimize chatbot flows, personalize offers, and fine-tune your marketing strategy.

To run this scenario, connect MCP using our detailed guide and confirm access to the chatbots_dialogs_list command.

Here’s a prompt example:

Analyze conversations of the bot_name chatbot and identify behavior patterns and insights in subscriber interactions. Recommend what we can improve based on these insights.

To start executing this task, your AI agent will ask for permission to access the conversations – click “Accept” to confirm it.

Within a minute or two, the assistant will send back a reply, including:

• the most common subscriber intents;
• peak activity times in incoming messages;
• friction points where users drop out of your pipeline or postpone target actions;
• conversion signals your chatbot or representative should react to.

Here’s what the reply looks like:

Your AI agent also suggests ways to enrich DM interactions and Instagram content overall, including automation ideas to lighten your team’s workload.

When your AI assistant uses the MCP server, it can accurately pinpoint communication issues and offer actionable tips that enhance user experience and team efficiency.

Create FAQs based on subscriber queries

Although AI agents can’t generate chatbot flows in SendPulse’s builder for now, they can come in handy when you create content. To give you an idea, you can ask it to analyze typical questions subscribers have, gather up-to-date information from your website, and generate concise answers to be used as chatbot auto-replies.

Here’s a prompt example:

Create flows with replies to the top FAQs based on user query analysis. When generating replies, use information from the official website site.com, including the following pages: site.com/about_us, site.com/delivery, site.com.com/returns/, and site.com/contact

Since your AI agent has already analyzed conversations, it doesn’t need any extra permissions or data. After a few seconds of reasoning, it delivers a list of suggested launch triggers for each FAQ flow, a list of flows, and content for each of them.

If the output chatbot flows don’t sound the way you want, tweak your prompt. Namely, upload examples of effective automations you already use or share your internal DM guidelines to help the AI agent follow your brand’s tone of voice.

For more granular control over the flow text, try breaking your prompts into smaller steps. You can start by asking AI to generate replies for the “Returns” flow. Once you’re happy with that, move on to other FAQ topics.

Launch chatbot campaigns

In your MCP settings, you’ll find five similar commands that start with chatbots_bots_campaigns. These let your AI agent create an actual chatbot campaign for Instagram, WhatsApp, Telegram, etc.

Available campaign commands vary by platform. On Instagram, your AI agent can:

• create a campaign with text, an image, and a file;
• send it to users active within 24 hours or schedule it for a specific date and time;
• segment recipients.

If you want to include extra elements, such as quick replies, buttons, or product cards, schedule your campaign through MCP first, then go to your SendPulse account and add these components manually.

Here’s a prompt example:

Schedule an “Event reminder” campaign for November 27, 2025, targeting subscribers of the bot_name chatbot with the event tag.

The message should remind them that November 28–30, we’ll host Stylist Weekend in our store in the heart of New York, featuring personalized sessions and discounts of up to 20%.

Subscribers have already received an email invitation, so the goal of this message is to remind them about the event, highlight its dates and time, and motivate them to visit our store.

Location: 45 Rockefeller Plaza, open from 11 AM to 10 PM.

In around a minute, your AI agent will schedule the campaign and confirm it in chat.

You’ll see it listed among scheduled campaigns in your SendPulse account.

Keep in mind that any follow-up actions on a scheduled campaign, such as editing or canceling, should be done manually. Say, you ask the AI agent to update the campaign’s text — it won’t warn you that this action isn’t supported. It will confirm the update but will instead create a new campaign for the same chatbot and audience segment using your updated content.

This means that both the original and the updated campaigns will be sent.

Since your AI agent cannot edit scheduled campaigns yet, any update requests will lead to duplicates. To avoid this, always edit campaigns manually or delete the ones you don’t need anymore first.

We recommend reviewing and refining your AI agent’s output before sending it out. As you know, AI can make mistakes, jump to wrong conclusions, or hallucinate information. Remember that AI is only a tool, which means the human using it is responsible for what ultimately gets sent.

Beyond scheduling, the AI agent can also suggest ways to boost your campaign’s performance. It might advise you to apply case-specific tags, warm up your target segment ahead of time so that more recipients fall within the 24-hour window, or ensure your account time zone matches recipients’ time zones.

You can also ask for more insights, like the optimal sending time, and AI will point you toward a data-backed window.

Segment users based on conversation patterns

This can be particularly helpful for marketers and business owners looking for advanced ways to segment audiences. If standard options like location, age, gender, or chatbot activity no longer give you valuable insights, try micro-segmentation based on user behavior and communication patterns.

Paired with our MCP server, AI can analyze a chosen number of conversations. It will look into how users communicate and what they talk about to uncover new segments for future messages. Your AI agent can also assign tags to contacts right in your SendPulse account to add a personal touch to your interactions.

Here’s a prompt example:

Analyze the bot_name audience and split subscribers into three groups based on their purchase intent. Assign the tags accordingly: ready to buy → hot tag; considering a purchase → warm tag; not ready to yet → cold tag. Give recommendations on how to communicate with each group effectively.

Before assigning tags, your AI agent goes over your conversations and groups subscribers based on your tagging criteria:

• hot ~10–15% — users who shared their delivery details or requested payment info, agreed to shipping conditions, or sent a payment confirmation;
• warm ~55–60% — active users asking about price, colors, sizes, or delivery terms and requesting extra photos or videos;
• cold ~20–25% — users who sent one-off or emotional reactions with no further interaction, raised objections like “it costs too much,” “you don’t have my size,” and similar, or stayed inactive for a long time.

AI also outlines behavioral signals for tagging and scoring rules and clarifies the setup steps.

Based on the prompt, your AI agent suggests how to interact with each segment to move leads down your sales pipeline and re-engage inactive users faster.

Once everything is confirmed, your AI agent starts assigning tags and shows real-time progress in the chat.

Test chatbot flows

Rather than analyzing your campaign statistics, your AI agent can create realistic simulations of your target audience to test chatbot flows before they go live.

Running the simulation will help you:

• avoid mistakes that real users come across in your flow;
• spot and fix any potential bottlenecks;
• predict click-through and open rates;
• strengthen the performance of your initial launch.

In this use case, AI adopts a virtual persona based on real conversation data, so it acts like a focus group that delivers instant and highly accurate feedback.

Here’s a prompt example:

Simulate 500 user interactions through the “Order Placement” flow to analyze it, spot bottlenecks, and predict how convenient the checkout process is for our target audience.

Flow entry point: 500 users who clicked “Place order” after interacting with ads or chatbot flows.

User steps: choose a payment method (cash on delivery or upfront payment) → send payment details and confirm the payment → after payment confirmation, send shipping details (full name, phone number, email address, and delivery address).

Based on it, AI quickly outlines key assumptions about user behavior, estimates that around 54% of subscribers will complete the flow in five to seven minutes, highlights friction points, and suggests how to boost conversions.

4 ways to use MCP with the CRM system

Standard CRM solutions store customer data points and help you manage them to build strong, long-term relationships. AI goes beyond that because it understands context. This means it can dig deeper into the data and act before issues surface. Case in point: by simulating real customer behavior, AI can pin down bottlenecks in your sales pipeline and predict potential churn.

To highlight how AI and MCP work together within SendPulse’s CRM system, we’ll take a pipeline of the same Instagram clothing store from the previous section.

Optimize a pipeline

Fine-tuning a sales pipeline usually takes time, money, and a team. It involves analyzing conversions, finding weak spots, adjusting sales scripts, running tests, and monitoring results.

AI lets you get instant insights without extra costs or staff. The MCP allows it to:

• access CRM data in real time;
• analyze thousands of deals in minutes;
• spot where exactly your business is losing money and deliver data-backed strategies to increase conversions.

AI moves past surface-level metrics, explains the reasons behind them, and offers solutions. This level of analysis goes far beyond a standard CRM report since it considers context, not just metrics.

To optimize your pipeline, connect the MCP server and grant it access to the crm_pipelines_list command.

Here’s a prompt example:

Analyze the ID 24759 pipeline and suggest actionable optimization steps to increase its overall efficiency.

The AI response includes a description of your pipeline and each stage, general tips for better performance, suggestions to enhance the conversion path, and in-depth recommendations.

Take this as an example: your AI agent may flag failed payments as an overlooked opportunity to boost conversions. This way, leads who have already made it from initial awareness to making a decision are more likely to complete their purchase.

To make it easier for sales reps to follow up on these leads, your AI agent adds a new stage for failed payments to your pipeline and highlights it in red.

To ensure the updated pipeline runs smoothly, AI creates a test deal and processes it through every stage, from a new lead to a successful deal. The test confirms that your pipeline is built out properly, and sales reps can use the “Failed payment” stage in their follow-up scripts.

AI can help optimize any element of your sales pipeline. It can reorganize stages, generate messages to include between stages, configure currencies, assign team members, and more.

Create deals after customer interactions

After each customer interaction, sales reps record key details in the CRM system to keep future communication on track. This may include contact names, phone numbers, deal amounts, stages, or tags, along with the conversation context. Without consistent notes and updates from reps, getting an accurate view of your sales pipeline becomes tough.

To create deals quicker, sales reps can pass essential information to your AI agent through MCP. AI will then independently process this information and add both the contact and the deal to your pipeline without anyone needing to interact with the CRM system firsthand.

Here’s a prompt example:

Create a new contact: Nick Williams, [phone number], [email address], ABCGroup company, Managing Partner.

Link this contact to a new deal in the “Online Store” pipeline, place it in the “Partnership” stage, set the due date to December 31, 2025, and add a deal value of $10,250.

AI uses a few tools, including crm_contacts_create, crm_deals_create, and crm_deals_update, and sends a reply in chat.

Although your AI agent can create a new contact, it can’t add the phone number, email address, job title, and other attributes due to the lack of these MCP tools. Instead, it notes these missing details in a comment.

For the same reason, the AI agent cannot add your deal’s due date and includes it in a comment instead.

This shows that AI won’t lose data even when certain tools are missing. It adapts and keeps all important information the best it can.

Our MCP server is still in its early stages of development, and we’ll remove these limitations as its features mature.

Create contacts and deals from chatbot conversations

In our earlier example, we’ve covered how AI agents create contacts and deals after a manual request from a sales rep. With the right approach, it can do more without substantial human involvement, like analyzing chatbot conversations, detecting completed orders, and automatically creating deals in your CRM system.

Here’s a prompt example:

Analyze the last 100 conversations of the bot_name chatbot. If a customer places an order in chat, create a deal in the “Online store” pipeline with the “Order placed” stage and “Product sale” type. Include all relevant details: client details, ordered items, price, delivery method, and payment method.

Consider this message a trigger for order placement:

“Please share your shipping details, including your full name, city, street, ZIP code, phone number, and email address.”

This isn’t a random trigger — your chatbot or sales rep actually sends this message once they receive a payment confirmation from customers. Once the payment goes through, the order counts as complete.

Keep in mind that AI cannot process images, so some steps may need human control anyway. To choose the right trigger, analyze your order flow. For instance, if it is fully automated, AI can create deals for every user who reaches the final element of your flow.

Once AI gets your prompt, it will generate a step-by-step plan to get you started.

Once approved, AI processes data and executes commands using the MCP server. Here’s what it does step by step:

• analyzes all conversations and filters those containing the trigger message;
• creates the first contact and adds a comment with the phone number and email address;
• creates a deal in the “Online store” pipeline with the “Order placed” stage, assigns a team member, and sets a deal type;
• links the contact to the deal;
• adds the deal amount and key attributes as a comment.

The whole process takes about 10 minutes.

When performing multi-step tasks, the MCP server may temporarily block actions if the AI model reaches the request limit. However, it doesn’t affect the output. Just wait for the amount of time shown in the notification, which can be from 1 to 40 seconds, and type in “Continue” in the chat.

Single out top-performing lead sources

A CRM dashboard shows where your leads come from, but it doesn’t reveal deeper insights like which sources actually drive revenue, how long deals take to finalize, or how engaged your team members are. Without a deeper analysis of this kind, it’s hard to distinguish lead quantity from lead quality. AI can help you with that, too.

Instead of simply counting conversions, it tracks the entire lead lifecycle and offers a clear picture of which communication channels bring customers who buy faster, spend more, and stay for longer.

Let’s say your AI agent has found that Facebook Ads bring only half as many leads as Google Ads, yet the average order value from Facebook is three times higher. With this insight, your marketing team can adjust budgets relying on actual business performance, not only gut feeling.

Here’s a prompt example:

Determine which lead sources generate the highest-quality leads, meaning those that become paying customers and move all the way to the “Deal Won” stage in the “Online Store” pipeline.

Give recommendations on how to improve lead-to-customer conversion rates.

Your AI agent requests access to deals and, within a few minutes, delivers a comprehensive reply that includes:

• a list of analyzed pipelines;
• communication channels that produce the best leads;
• recommendations that boost lead-to-customer conversions;
• a list of metrics that lead to these conclusions.

AI found chatbots to be the best source of high-quality leads — business data proves it true. Suggestions were focused on how to move leads through the pipeline faster and enhance the accuracy of deal fields in general.

A data-rich pipeline helps your AI agent run deeper analysis and offer stronger insights. Beyond revenue from each lead source, AI can also consider other metrics, such as retention rate, sales cycle length, average order value, failed payments, and more.

Building more efficient AI agents

We believe MCP servers are the future of how we’ll all interact with AI. SendPulse’s MCP server gives AI real-time access to your data sources and the ability to execute straightforward commands. And this is just the beginning of how we’re using this technology. Next, we’ll see the transition to self-running AI agents that can handle complex workflows and take action on their own.

Future AI agents will understand broader context across all communication channels, including chatbots, web push notifications, emails, and SMS campaigns. They’ll track pipeline changes, analyze customer behavior, and proactively suggest new interaction strategies. AI will also simplify CRM workflows by automating almost every step. In this new environment, humans shift from executors to AI ecosystem curators, setting goals rather than managing individual tasks.

Ready to get started? Connect the MCP server to create your own AI agent and discover how it frees up time, boosts day-to-day efficiency, and accelerates business performance.