AI assistants such as Claude and ChatGPT have become part of how many people work. They draft emails, summarise documents and answer questions. What they usually cannot do is see your business: who called yesterday, what a customer asked for, which appointments are booked this week, which follow-ups are overdue. Without that context, every answer starts with you copying and pasting.
The Cypher LM MCP server closes that gap. It lets an AI assistant connect to your Cypher LM account through the Model Context Protocol, an open standard for connecting AI tools to other systems, so the assistant can read your calls, check your bookings, look up a customer's history, list open follow-ups, and start task calls on your behalf.
This guide explains what MCP is, what the Cypher LM MCP server can and cannot do, how to connect an assistant to it, how to keep it safe, and some practical ways to use it. One note before we start: the Cypher LM developer platform, which includes the MCP server, the REST API and webhooks, is built and currently in beta. We are opening it to accounts gradually; if you would like access, contact our team at team@cypherlm.com.
What the Model Context Protocol is
The Model Context Protocol, usually shortened to MCP, is an open standard that describes how an AI assistant can discover and use tools provided by another system. A system that offers tools is called an MCP server; an assistant that uses them is an MCP client.
The idea is simple. Instead of every AI product building a custom integration with every business tool, each tool publishes an MCP server describing what it can do, such as "list recent calls" or "get a customer's history", and any MCP-compatible assistant can use it. When you ask the assistant a question, it decides which tools to call, calls them with the right inputs, reads the results and answers you in plain language.
For you, that means you can ask questions about your business in natural language and get answers based on your real data. "Which customers called about pricing this week?" becomes a question the assistant can answer by reading your calls, rather than one you have to answer by scrolling through them.
MCP does not give an assistant unlimited access. It can only use the tools the server offers, with the access granted by the credentials you provide. In Cypher LM's case, that means your API key, which is tied to your account and can be revoked at any time.
What the Cypher LM MCP server can do
The Cypher LM MCP server offers eight tools. Each maps to something you can already see or do in the dashboard, and each has a clear purpose.
- list_agents: lists your voice agents with their name, status and purpose. Useful as a first step when you run several agents.
- list_recent_calls: lists recent calls, newest first, optionally for one agent. Test calls are excluded. Ask "What calls came in this morning?"
- get_call: returns one call in full, including its transcript. Ask "What exactly did the caller say about the delivery date?"
- get_customer_history: returns one customer's history across calls, bookings, meetings, task calls and open follow-ups, by phone number or email. Ask "What has Emily Carter been in touch about this month?"
- list_bookings: lists bookings, upcoming by default, with optional date range and status. Ask "What appointments do we have on Thursday?"
- list_follow_ups: lists follow-ups agreed on calls and in meetings, such as open items. Ask "Which follow-ups are overdue?"
- create_task_call: starts a task call, in which one of your agents contacts a customer with a goal and questions to ask. Ask "Call James Miller to confirm his delivery address and ask for the best time to deliver."
- get_task: returns a task call's status and outcome, including the result, a summary, the answers collected and next steps. Ask "What did James say?"
Seven of the eight tools read information. The eighth, create_task_call, takes an action: it starts a call, which uses your agent's minutes. Task calls reach the customer by phone if a number is connected to your account, or with a personal call link sent to them by email if not.
The tools are designed to be combined. A good assistant will chain them on its own: to answer "Is there anything I should know before my call with Sarah Lewis?", it might look up her history, open the most recent call for detail, and check her upcoming bookings and open follow-ups, then give you a short brief. You ask one question; the assistant does the lookups.
Because the answers come from your records, they are only as complete as those records. Customers who share a phone number or email on their calls are grouped into one history, while callers who share neither cannot be looked up by name. If an answer seems thin, the underlying calls usually explain why.
What it does not do
Being clear about the limits is as important as describing the tools. The MCP server is designed for reading your business data and starting task calls, not for managing your account.
It cannot create, move or cancel bookings. It cannot change your availability, hours or appointment types. It cannot create or edit agents, change their knowledge, or alter any settings. It cannot send payment requests or refunds. Those actions stay in the dashboard, where your team can see and confirm them.
That boundary is deliberate. Reading data is low risk and high value: an assistant that can summarise your week's calls or prepare you for a customer conversation saves real time. Changing your calendar or your money flows on the basis of a misread instruction would be a different kind of risk. If you need those actions automated, the REST API and webhooks are the place to build them with your own checks.
Task calls are the one action, and they deserve care. Before your assistant starts one, check the goal and the customer details it has chosen. A good habit is to ask the assistant to tell you what it is about to do before it does it.
| Option | Best for | How it works | Who uses it |
|---|---|---|---|
| MCP server | Asking questions and starting task calls in plain language | An AI assistant calls Cypher LM tools for you | Owners, managers, anyone using an AI assistant |
| REST API | Building your own integrations and reports | Your software calls Cypher LM endpoints | Developers |
| Webhooks | Reacting when something happens | Cypher LM sends signed events to your URL | Developers, automation tools |
In practice, the split is about who is asking and how often. The MCP server suits people who want answers on demand and do not write code. The REST API suits developers building something that runs on its own, such as a nightly export. Webhooks suit anything that should happen the moment an event occurs. Many businesses end up using the MCP server for people and webhooks for systems.
Connecting Claude, ChatGPT or another MCP client
Connecting an assistant takes a few minutes once developer access is enabled for your account.
- Create an API key. In Cypher LM, go to Automations › Webhooks & API keys and choose Create key. Give it a name that says where it will be used, such as "Claude — Sarah's laptop". Copy the key straight away; it is shown only once.
- Copy the MCP server address. The Endpoints section on the same page shows the MCP server URL, https://api.cypherlm.com/api/v1/mcp.
- Add it to your MCP client. In your assistant's settings, add a custom MCP server or connector of the Streamable HTTP type, using the URL above, and set the API key as the Bearer token for authorisation.
- Try a simple question. Ask "List my Cypher LM agents" or "What calls came in today?" to confirm the connection works.
Claude supports connecting to remote MCP servers as custom connectors. ChatGPT can connect to remote MCP servers where your ChatGPT plan supports custom connectors. Other MCP-compatible clients, including developer tools, work in the same way. The exact menu names differ between clients and change over time, so follow your client's current instructions for adding a remote MCP server with a bearer token.
If the connection fails, the most common causes are a mistyped key, a revoked key, or an account whose plan does not include the API. The server replies with a clear error in each case.
Writing good requests
Assistants work best with specific requests. "What happened yesterday?" will produce a general summary; "List yesterday's calls where the caller asked about prices, with the caller's name and what they asked" will produce something you can act on. Name the customer, the time period and the output you want, and the assistant will choose the right tools.
For task calls, be explicit about the goal, the customer and the answers you need, just as you would in the dashboard. "Call James Miller on his mobile to confirm Thursday's 2 pm installation and ask whether parking is available" gives the agent everything it needs. Ask the assistant to read the request back before it starts the call, and check the outcome afterwards with get_task.
Keeping it safe
An API key is a password for your Cypher LM data, and it should be treated like one. Anyone who has your key can read your calls, including transcripts, and start task calls.
Create a separate key for each assistant or person, named so you know where it is used. That way, if one is lost or no longer needed, you can revoke it without affecting the others. Revoking a key takes effect immediately, and anything using it stops working straight away. Keys are stored by Cypher LM only as a secure hash, which is why they are shown just once.
Each key can make up to 120 requests a minute. That is plenty for an assistant answering your questions, and it limits the impact of a misbehaving script. If the limit is reached, requests are refused briefly and the client is told when to retry.
Remember that the assistant itself sees what it reads. When you ask it about a call, the transcript is sent to the assistant you are using, under that assistant's own terms. Choose an assistant and plan whose data handling you are comfortable with, especially for sensitive conversations.
How this works in Cypher LM
The MCP server is one part of the Cypher LM developer platform, alongside the REST API and webhooks. All three share the same API keys, created in Automations › Webhooks & API keys, and the same plan rules: the developer platform is available on Trial, Premium and Enterprise plans.
The data the MCP server reads is the same data your team sees in the dashboard. Calls come from your Cypher V agents, with recaps from Cypher N. Bookings come from Cypher S. Customer history comes from Cypher C, combining calls, bookings, meetings, task calls and follow-ups. Follow-ups come from call recaps and meeting notes. Task calls use the same Calls & tasks feature your team can use from the dashboard.
That means the assistant's answers are only as good as the records behind them. Agents with clear knowledge, bookings made through Cypher S, and customers identified by phone or email all make the assistant more useful. Developers who want to build their own integrations can read our guide to the Cypher LM APIs.
Practical uses tend to fall into three groups. Morning briefings: "Summarise yesterday's calls and list anything that needs a call back." Preparation: "I'm about to call Sarah Lewis — what does she want, and is anything still open?" Follow-through: "Which follow-ups are overdue, and can you start a task call to confirm next week's appointment with the customers who have not confirmed?"
Most businesses start with read-only habits and add task calls once they trust the results. A sensible first week is to use the assistant only for questions: morning summaries, customer preparation and follow-up reviews. Compare its answers with what you see in the dashboard. Once you are confident in the answers, start using it to launch task calls for routine confirmations.
Frequently asked questions
Is the MCP server available on my account?
The developer platform, including the MCP server, is in beta and being opened to accounts gradually. It is included on Trial, Premium and Enterprise plans. Contact team@cypherlm.com to request access.
Can the assistant book or cancel appointments?
No. The MCP server can list bookings but cannot create, move or cancel them, or change your availability. Those actions stay in the dashboard or on calls with your agent.
Does starting a task call cost anything?
Task calls use your agent's minutes, like any other call. Review the goal and customer details before your assistant starts one.
Which AI assistants can connect?
Any MCP client that supports remote servers over Streamable HTTP with a bearer token. That includes Claude, and ChatGPT where your plan supports custom connectors.
What happens if a key is leaked?
Revoke it in Automations › Webhooks & API keys. Revocation takes effect immediately, and anything using the key stops working.
Conclusion
The Cypher LM MCP server turns your AI assistant into something that knows your business: who called, what they wanted, what is booked and what is still open. It reads your real records, answers in plain language, and can start task calls when you ask, while leaving bookings, settings and payments in your hands.
If you already use Claude or another MCP-compatible assistant, it is one of the quickest ways to get more from the calls your agents handle. To request developer access, write to team@cypherlm.com, and if you are new to Cypher LM, request a free trial to start building the records your assistant will read.
