Integrations
MCP connectors: AI works with your systems and your contact center
MCP (Model Context Protocol) is an open protocol that lets AI connect to data and actions in other systems. CONTACT.WORK uses it both ways. Our AI agents use MCP connectors to look up data in your CRM, ERP, or booking system and take action without a person: book an appointment, create a deal, send an invoice. And the CONTACT.WORK MCP server opens your contact center to your own AI – Claude, ChatGPT, or an agent your team built. You can ask it in plain English how calls and inquiries went, and hand it tasks.
Capabilities
What you get with MCP connectors
Examples
What AI does through MCP
Three typical jobs: an AI agent resolves a customer's issue, a manager makes sense of the numbers, and a team hands routine work to AI.
Customer issue resolved on the call
A customer asks to move an appointment. The voice agent finds them in your booking system, offers open slots, reschedules, and sends a confirmation by text or messaging app. No agent needed, and the CRM gets a note about the call.
- Open slots from your booking system
- Rescheduling without an agent
- Note in the CRM
yclients.find_clientMary Smith, visit Wednesday 11:00 AMfoundyclients.free_slotsFri 12:00 PM, Fri 4:30 PM, Sat 10:00 AM3 slotsyclients.movevisit moved to Friday, 4:30 PMdonemax.sendconfirmation to customer in MAXsentcrm.notecall note in CRMsavedAnswers from contact center data
A manager asks their AI: "Why did we miss more calls yesterday?" The AI pulls shift and queue metrics from CONTACT.WORK, finds the peak hour and the agents on break, and answers with specific calls.
- Queues, shifts, and missed calls
- Inquiry topics from transcripts
- Links to specific calls
Tasks in your own words
"Call customers with unpaid orders" or "Message everyone booked for tomorrow" – the AI builds the list and starts the outbound campaign or the messages. Before launch, it shows the list and waits for a person to approve it.
- Outbound calls from a list
- Messaging app and SMS messages
- Launch after approval
1c.unpaid41 orders unpaid for over 3 daysfoundList41 customers, amounts and due datesshownConfirmationKaren Miller, 11:20 AMacceptedcalls.campaignoutbound calls to 41 customersin progressWe set up complex workflows on request, such as an outbound campaign with different logic for each customer.
Catalog
Prebuilt MCP connectors and the MCP server
CRM
Bitrix24, Kommo (amoCRM), and other CRMs
From the phone number, the agent identifies the customer, sees their deals and orders, creates a deal or task, and logs the call outcome. Sales reps get everything in their CRM with no manual copying.
- Customer and deals by phone number
- New deal or task
- Call outcome on the record
amocrm.find+1 (555) 010-0115: deal “Supply, batch 2”foundamocrm.noteoutcome: agreed on a 5% discountloggedamocrm.taskMike – send the contract by FridaycreatedERP and booking
ERP, inventory, and booking systems
Order status, stock and prices, customer balance, a specialist's open time – the agent pulls it from your system the moment the customer asks. On request, it sends an invoice, books an appointment, or reschedules a visit.
- Orders, stock, balances
- Invoice on request
- Booking and rescheduling
1c.order#4518 packed, ships tomorrowreplymoysklad.stock“Nord” armchair, gray – 7 pcsin stock1c.invoice$1,840 invoice sent by emaildoneKnowledge and tasks
Confluence, Google Drive, cloud storage, task trackers
The agent answers from your policies, price lists, and instructions, not from the model's general knowledge. If a question needs a person, it creates a task and tells the customer when to expect an answer.
- Answers from your documents
- Link to the answer's source
- Task for an employee
confluence.searchReturns policy, section 4.2foundReplyyes, with proof of paymentto customerplanfix.taskcheck return #4518, due tomorrow 6:00 PMcreatedYour own systems
Databases and REST services
In-house systems connect too: the agent reads a PostgreSQL or MySQL database or works with a REST service from its OpenAPI description. Writing to the database and complex action chains are available on request.
- Read access to PostgreSQL and MySQL
- REST service via OpenAPI
- Database writes on request
PostgreSQLorders, read-onlyconnectedMySQLtickets, read-onlyconnectedOpenAPIpartner portal, 12 methodsconnectedBookingdatabase changeson requestYour AI
CONTACT.WORK MCP server
Connect the contact center to Claude, ChatGPT, or an agent your developers built. The AI sees queues, agents, and missed calls, finds calls and chats with transcripts, messages customers, starts outbound campaigns, and assigns callbacks to agents.
- Queue and agent metrics
- Calls, chats, transcripts
- Messages, campaigns, tasks
queue_statsqueues, wait time, and abandonment by hourfind_callscalls and chats with transcriptssend_messagemessage to customer in MAX, Telegram, SMSstart_campaignoutbound calls or broadcast from a listcallback_taskcallback task for an agentControl
Permissions, approvals, and log
Every action has its own permissions: the AI may only read, or read and change. Refunds, order cancellations, and other irreversible steps run only after a person approves. Every action goes into the log: who asked, what the AI did, and what the system returned.
- Read or change
- Approval for irreversible steps
- Log of every action
find_clientread-onlyallowedmove_visitdata changesallowedrefundafter human approvalwaitingdelete_clientclosed to AInot allowedLog214 calls today, 1 awaiting decisionopenSetup
How we connect MCP connectors
- 01TasksWe decide which tasks AI handles by itself and where a person is needed.
- 02AccessYour admin creates accounts in your CRM, ERP, or other system.
- 03PermissionsFor each action we set: read, change, or change with approval.
- 04TestingWe test the agent on sample conversations and check every action against the log.
- 05Your AIWe issue an access key to our MCP server and connect it to your AI.
- 06LaunchWe turn the connectors on and review the log with you for the first few weeks.
Requirements
What you need for MCP connectors
- Hosting
- Wherever the platform runs: our server or yours
- Model
- Any model that supports tool calling, including a local one
- Your AI
- Claude, ChatGPT, Cursor, and other MCP-enabled apps
- System access
- An account or API key with the right permissions
- Log
- All AI actions are logged in the platform and available for review
- Cost
- A prebuilt connector – in support hours; a new one – in development hours
Preparation
What to prepare for MCP connectors
Q&A
Frequently asked questions
How is an MCP connector different from an API integration?
An API integration links two systems in advance: an event in one triggers an action in the other. An MCP connector gives AI a set of tools, and the AI picks the right one during a conversation or question. Inside, a connector usually uses the same APIs – it's just the AI calling them.
How do we connect our AI to the contact center?
We issue an access key, and your admin adds the MCP server to your AI's settings. From there, the AI sees which tools it has and uses them when a question calls for it.
What if the AI makes a mistake?
It's easy to find: the log records every AI step along with your system's response. You can narrow permissions for that action right away – for example, read-only until you find the cause.
Will customer data go to an outside AI?
The access key decides what your AI sees: you can give it only aggregate metrics and hide customer names and phone numbers. If the AI needs full conversations, use a model deployed on your servers – then the data stays with you.
Which of your agents use connectors?
All of them: the voice AI agent on the line, the chatbot in messaging apps, and the assistant that suggests replies to agents during calls. You connect a connector once, and each agent gets access under its own permissions.
Our system isn't on the list. What now?
If it has an API or a database, we build a connector for it. That's custom work, estimated in development hours.
How much do MCP connectors cost?
Connecting a prebuilt connector is billed in support hours under your support plan. A new connector is built as development work: Maxi plan customers use hours already included in their plan, and on the Pro plan it's billed separately at $60 per hour.
Glossary
Terms on this page
- Connector
- A prebuilt module that links the platform with telephony, a CRM, or another system. Learn more →
- MCP (Model Context Protocol)
- An open protocol that lets AI work with the data and actions of other systems. Learn more →
- CRM – customer relationship management
- Where a company keeps its customers, deals, and full communication history. Learn more →
- ERP – enterprise resource planning
- Orders, inventory, and finance in one system. Learn more →
- API (application programming interface)
- The way programs exchange data with each other without a person in the loop. Learn more →
- Booking system
- Software where travel companies manage booking requests and bookings. Learn more →
- REST: exchanging data between systems over HTTP
- The standard way software exchanges data over the internet. Learn more →
- Chatbot
- Answers customers in chat and messaging apps from a scripted flow with buttons. Learn more →
- Agent assist
- Suggests answers from the knowledge base to the agent during a conversation. Learn more →
- Cloud deployment (cloud)
- The platform runs on the vendor's servers, and agents connect over the internet. Learn more →
- Callback
- The customer doesn't wait on hold but asks for a callback – and gets one when an agent is free. Learn more →
- Hosting
- You rent a server from a provider instead of buying one. Learn more →
- Knowledge base
- All answers, policies, and prices in one place – for agents and AI. Learn more →
- Large language model (LLM)
- A neural network that understands questions asked in everyday words and answers the same way. Learn more →
Updated: