We want to show some examples so you can see our data is accurate, and to show some interesting MCP servers.
Payments · San Jose, CA · MCP Server
PayPal is one of the first big payment companies to put its services behind an MCP server, which means small business owners can now handle PayPal tasks just by chatting with an AI assistant instead of clicking around inside the PayPal website.
Start with invoicing. A merchant can open up an AI chat tool and type something like "create an invoice for landscaping services for Green Lawns for $200 dated last Friday," and the assistant talks to PayPal in the background and produces a real, sendable invoice link. No logging into a dashboard, no filling out forms.
It gets more useful when the AI assistant is also connected to the merchant's files. Someone can point it at a spreadsheet of clients and ask it to send out a batch of invoices in one go, and the assistant pulls each customer's details and creates the invoices one after another.
Invoicing was just the starting point. PayPal has since opened up the full set of tools a merchant might want from an assistant, including creating payment orders, capturing payments, pulling transaction history, and issuing payouts to contractors or employees.
To handle the load of running all this at PayPal's scale, the hosted version of the server runs on Cloudflare. That lets PayPal push the AI assistant traffic out to servers close to wherever the merchant is in the world, so requests like "send these ten invoices" come back quickly instead of stalling.
The end result is that the AI tools merchants already use, including Claude, ChatGPT, and the Windsurf code editor, can now log into PayPal on their behalf with a simple click and start getting real work done.
Communications · San Francisco, CA · MCP Server
Twilio built its MCP server for a different audience than most companies. Instead of helping end users send texts through a chat window, the server is aimed squarely at developers writing code, giving their AI coding assistant a way to look up exactly how Twilio's APIs work without leaving the code editor.
The problem it solves is familiar to anyone who has tried to build with Twilio. The company has more than 1,800 different API endpoints across 30-plus products, covering everything from sending a text message to running a video call to routing customer service tickets. An AI coding assistant trying to write code against all of that often gets confused, uses the wrong endpoint, or skips a required step.
The server fixes this with two simple tools. The assistant can search across all of Twilio's documentation in plain English ("how do I send an SMS?") and pull back the exact, up-to-date specification for whatever API call it needs, including every parameter. So when a developer asks their assistant to "set up sentiment analysis on customer support calls with automatic escalation to a human," the assistant can actually find the right three Twilio products and wire them together correctly.
This is especially useful for Twilio's newer products. Those products aren't in the training data that AI assistants learned from, so without the MCP server the assistant would have no idea they exist.
To run all of this at scale, Twilio hosts the server on AWS. The server is read-only and indexes only public documentation, so developers don't need a Twilio account or login to use it. They just point their coding assistant at a single URL.
The result is that any developer using Claude Code, Cursor, Codex, or a similar tool can get accurate Twilio code suggestions on the first try, instead of going back and forth between their editor and a browser tab full of documentation.
Tax Compliance · Durham, NC · MCP Server
Avalara handles the unglamorous but necessary work of tax compliance, calculating sales tax for online orders, filing tax returns, managing exemption certificates, and so on. Their MCP server lets businesses hand all that work over to an AI assistant instead of clicking through dashboards or writing custom code.
What makes Avalara's approach stand out is that they didn't build just one server. They built a whole family of them, with each one covering a different piece of compliance. There's one for calculating taxes on a sale, one for filing returns, one for sending electronic invoices, one for tracking business licenses across states, and one for managing the exemption certificates that nonprofits and resellers use to avoid paying tax.
A practical example looks like this. A finance team can ask their AI assistant to "calculate the sales tax on this order to a customer in Texas, then file it on next month's return." The assistant uses one server to do the calculation and another to make sure it ends up on the right return, without anyone touching a spreadsheet.
The split is deliberate. Different teams at a company care about different things, and an AI assistant only sees the slice it has permission to use. The accounts payable group doesn't need access to business license renewals, and an outside developer building a checkout integration only needs the tax calculation piece.
To run all of this, Avalara hosts the servers on AWS, which lets them keep up with the heavy traffic that comes from tax calculations happening on every order across every customer they serve. Avalara has been a longtime AWS customer, and putting the MCP servers on the same infrastructure means the AI requests get the same uptime and scale as the rest of their platform.
The end result is that an AI assistant can handle the parts of tax work that used to require either a specialist or a lot of clicking, from quoting the right tax on a sale to filing the paperwork that proves the company stayed compliant.
Travel Technology · Southlake, TX · MCP Server
Sabre runs much of the plumbing behind airline and travel agency bookings, and they were the first company in travel to build an MCP server for it. The pitch is that any AI assistant can now do the kind of complicated travel work that used to require sitting on hold with an airline or clicking through a maze of agent screens.
The use cases get specific fast. An AI assistant connected to Sabre could wait on hold with an airline after a flight gets canceled, rebook the traveler on the next available flight, pay with a stored card, and update their calendar, all in the background while the person goes about their day. Another could check a traveler into a hotel at midnight, hold the room so it doesn't get given away, and arrange oat milk for breakfast.
There's also the agent-to-agent angle. One travel agency's AI can talk directly to another agency's AI to work out the messy parts of a booking, like splitting a ticket across two airlines or figuring out which fares can be combined. That sort of back-and-forth used to mean phone calls or emails between human agents.
The server also handles the paperwork side of travel. An AI assistant can fill out a visa application, pay the fees, and attach the documents to the booking, or gather up receipts from a trip, code them to the right expense categories, and file a complete report that follows the company's policy.
What makes this work is that Sabre plugged the MCP server into the same platform that already powers their airline and agency customers, so the AI workflows get the same scale and reliability as the rest of the system. Travelers and agents don't have to wait hours for things that used to eat up an afternoon.
Payments · South San Francisco, CA · MCP Server
Stripe was one of the first companies to ship an MCP server, and the most interesting thing about it is what they're letting AI assistants do: actually move money. Their server lets an AI agent create customers, charge a card, issue refunds, send invoices, and manage subscriptions on a real Stripe account.
A simple use case looks like this. A founder running a small SaaS company can ask Claude or Cursor to "create a new customer for jane@example.com, set up a $49 monthly subscription, and send her the first invoice." The assistant talks to Stripe through the MCP server and gets it all done without the founder touching the Stripe dashboard.
To keep this safe, Stripe layered in real guardrails. Permissions are controlled by a restricted API key, so a developer can give an assistant the power to read customers but not refund anything, or to create products but not delete them. There's also a manual confirmation step before the assistant fires off anything that moves money for real.
The server also indexes Stripe's own documentation. So when a developer asks "how do I handle a failed webhook signature?", the assistant can search through Stripe's docs and pull the right answer, instead of inventing code that looks plausible but doesn't actually work.
Stripe is also using the server as a starting point for something bigger. They've been building a stack of related protocols for agents to make payments on behalf of users, so that an AI shopping for hiking boots, for example, can find the right pair, hand over a one-time virtual card, and complete the purchase end-to-end with the usual fraud protection and receipts in place.
The end result is that small companies and solo developers can run a real payments business by talking to an AI assistant, and bigger companies can start building products where the AI itself is the customer paying for things.
Investment Research · Chicago, IL · MCP Server
Morningstar is best known for the star ratings it slaps on mutual funds and the deep research it publishes on stocks, ETFs, and retirement strategies. Their MCP server lets a financial advisor or investor pull all of that into an AI chat instead of clicking through Morningstar's website to find it.
The server keeps things focused with two tools. The first one looks up specific numbers about an investment, things like a stock's fair value estimate, its Morningstar rating, its market cap, earnings per share, or for a mutual fund, its net asset value and total return. The second one searches Morningstar's library of editorial research, the articles where their analysts explain what they think about a company, a sector, or a strategy.
In practice this means an advisor can ask an AI assistant something like "what's Morningstar's fair value estimate on Apple and what do their analysts say about the long-term outlook?" The assistant grabs the number from the first tool and the analysis from the second, and answers in one shot. No tabs to juggle, no separate logins.
The split between the two tools is deliberate. Morningstar's value comes from two very different things: the cold, hard data points that they're known for, and the opinions and methodology that their analyst team produces. Mixing them in one tool would make it harder for the AI to know which one to reach for, and harder for the user to tell where an answer came from.
Among major financial data vendors, Morningstar has gone further than most. Bloomberg, LSEG, and FactSet have all said they're exploring MCP but haven't released a public server, partly because their data carries strict licensing rules. Morningstar's ratings and research are licensed differently, which made it easier for them to be the first big name in this space to put up a server that paying customers can connect to from their own AI tools.
Global Employment · San Francisco, CA · MCP Server
Remote.com is the company that handles the paperwork side of hiring people in different countries: the contracts, the payroll, the local tax rules, the compliance bits that change every time you cross a border. Their MCP server lets HR teams ask questions about all of that in plain English instead of digging through spreadsheets or building reports.
The use cases get specific fast. An HR leader can open up Claude or ChatGPT and ask "pull comp data for the product team and flag anyone who hasn't had a salary adjustment in 18 months and who's below the country midpoint." The AI assistant pulls the numbers from Remote's live data, compares them to country-specific salary benchmarks, and comes back with a ranked list of who needs a raise.
Compliance work is the other big use case. A legal team needs a snapshot of every employee in Germany and France, with their contract expiry dates, probation status, notice periods, and contract types, formatted as a table. That kind of request used to mean an export, a few VLOOKUPs, and an hour of fiddling. With the MCP server, the assistant pulls it directly from Remote and hands back the table.
There's also a self-service angle for individual employees. Someone can ask their AI assistant "how many days of PTO do I have left, and when does my contract end?" and get an answer scoped only to their own record. A manager asking the same question across their team gets a different answer, because their permissions in Remote let them see more.
The security setup is what makes the whole thing work for HR data. There are no API keys to generate or store. An employee signs in through their browser the same way they log into Remote, and whatever permissions they already have carry over. An admin sees the whole company, an employee sees only themselves, and the AI assistant can't pull a raw export of the database. Remote sends only what was asked for, and the workforce data isn't used to train any AI model.
Fraud Prevention · New York, NY · MCP Server
Forter is a fraud-prevention company that sits in the background of major retailers like Nordstrom, Instacart, and Priceline, deciding in real time whether a checkout looks legit or looks like fraud. Their MCP server, part of a product they call Trusted Agentic Commerce, solves a brand new problem: telling the difference between a real customer's AI shopping agent and a scammer's bot.
The problem showed up almost overnight. When OpenAI released its ChatGPT shopping agent, Forter saw agent-driven traffic on the sites it protects jump by more than 18,000% in a single day. Six months later, that number kept climbing, with agent activity up over 2,000% across its network. Merchants suddenly had a new question: when a bot adds a $400 pair of headphones to a cart and tries to check out, is that an actual person's assistant doing the shopping for them, or is it a fraudster running a stolen credit card through a script?
The traditional answer used to be: just block all bots. Forter's data shows that's exactly what about 40% of travel sites did. But that approach also blocks the customer who delegated a real purchase to Claude or ChatGPT, which means the merchant loses the sale to a competitor who let the agent through.
Forter's MCP server gives merchants a third option. When an AI agent shows up at checkout, the merchant's site can ask Forter, through the MCP server, whether the agent is acting on behalf of a real person with a real history, or whether it's just an anonymous bot trying to test stolen cards. Forter compares the request against its identity network, which has handled over a trillion dollars in transactions, and gives back a trust decision in real time.
The setup is sold through the AWS Marketplace as a plug-and-play piece that merchants can drop into the parts of their checkout where an AI agent might show up. The MCP server is the part that lets the merchant's own AI workflows ask Forter questions in real time, which means as agentic shopping grows, Forter becomes the trust layer underneath it.
CRM · Cambridge, MA · MCP Server
HubSpot built two different MCP servers, which is a sensible move once you think about who actually uses HubSpot. There's the sales rep and marketer working inside the CRM every day, and there's the developer building custom tools and apps on top of HubSpot's platform. The two groups need very different things from an AI assistant, so HubSpot gave each one its own server.
The first one lets a salesperson or marketer talk to their CRM in plain English. Someone can ask Claude or ChatGPT to "summarize all deals in the 'Decision maker bought in' stage with a value over $1,000," or "pull the last five support tickets for Alex Smith." The assistant grabs the data from HubSpot in real time and answers the question, with no spreadsheet exports or hunting through dashboards involved.
The same server can also update records based on what someone tells the AI assistant. After a client meeting, a rep can say "update John Smith's contact with his new email and note that he's now Director of Operations," and the change shows up in the CRM. The assistant covers contacts, companies, deals, tickets, products, orders, invoices, quotes, and subscriptions, which is most of what a typical sales or service team works with day to day.
The second server is for developers building on HubSpot's platform. Instead of giving the AI assistant access to CRM data, this one gives it access to HubSpot's developer documentation and command-line tools. A developer can ask their AI coding assistant "what's the name of the component for displaying a table in HubSpot UI Extensions?" and get the right answer pulled from HubSpot's actual docs, instead of code that looks plausible but uses an outdated function name.
The split is the smart part. HubSpot's data scopes are tied to a user's permissions in the CRM, so a sales rep's AI assistant can only see what the rep is allowed to see. A developer's AI assistant doesn't need to see customer data at all, just the docs and the tools to build new features. By keeping the two flows separate, HubSpot avoided the problem of one giant server trying to be everything to everyone.
Revenue Orchestration · Atlanta, GA · MCP Server
Salesloft and Clari merged not long ago, and one of the first things the combined company shipped is an MCP server that pulls together their two halves of the sales process: Clari's view of which deals are likely to close, and Salesloft's view of what reps are actually doing to close them.
The use case the server unlocks looks like this. A sales manager can ask Claude or ChatGPT something like "show me the deals in this quarter's forecast that are slipping, and tell me which reps haven't followed up with the buyer in the last two weeks." The assistant pulls the forecast data from one side of the platform and the rep activity from the other, and answers in one shot. Before the merger and the MCP server, this would have meant logging into two separate tools and reconciling the data by hand.
The server also closes the loop between insight and action. If the assistant flags a stalled deal, a manager can ask it to draft a follow-up email and queue it in the rep's task list, without leaving the chat. Reps can do the same with their own conversations, turning the action items from a recorded sales call into actual tasks and emails to send.
What makes this work is that Salesloft already sits in the middle of how sales teams operate every day. The platform tracks every email, every call, every meeting a rep has with a buyer. By exposing all that activity through an MCP server, any AI tool a sales team already uses can suddenly answer questions that used to require a dedicated revenue operations analyst, like which buyer behaviors actually correlate with closed deals, or which sales plays are working in which industries.
The result is that a sales leader's AI assistant can sit on top of their entire revenue motion, from the forecast number their CFO cares about all the way down to the specific email a rep is about to send.
CSLLC
source
FINAM
source
WaLead.ai
source
ID Athletic
source
Plaud
source
WholesomeCo Cannabis
source
Little Engine
source
COROS
source
Reap
source
Adaptic
source
Tanka
source
Zwitch
source
CrowdReply
source
PropertyManagement.com
source
Lumonic
source
Classavo
source
preparedEPA
source
Carbon DeFi
source
Fraudlogix
source
PayPal
Twilio
Avalara
Sabre
Stripe
Morningstar
Remote.com
Forter
HubSpot
Salesloft
