AI CRM That Works Both Ways: You on the Board, Your Agent in Chat

A two-way AI CRM is one CRM with two equal ways in. You work on a normal pipeline board in the browser, and your AI agent (Claude or ChatGPT) works on the same data from chat through an MCP connector. What you drag on the board, the agent sees. What the agent changes in chat, you see on the board. And the agent can change more than records: it can reshape the CRM itself, with every setup change saved in a history you can roll back.

That is the idea behind the KMS AI CRM. This article explains why most "AI CRMs" only work one way, how the two sides fit together, what a real round trip looks like, who it suits, and where the limits are. If you want copy-ready prompts, see how to run your CRM from Claude. If you want a side-by-side of tools, see our comparison of AI CRMs.

Why most "AI CRMs" are one-way

Most tools sold as an AI-powered CRM follow the same pattern. There is a classic CRM interface, and somewhere inside it there is an AI panel. It can summarize a deal, draft a follow-up email, or answer a question about the records on screen. Useful, but the AI lives inside the app and works for the app.

That makes it one-way in two senses:

• You have to come to it. The assistant only exists inside that product. If you spend your day in Claude or ChatGPT, the CRM's AI is somewhere else.
• It touches records, not the system. It can read and maybe edit deals, but the pipeline stages, fields and rules are still set up by a person clicking through settings screens.

So when your process changes, you stop, open settings, and rebuild things by hand. The AI watches.

The two sides of a two-way AI CRM

The human side: a normal CRM in the browser

At crm.kissmyskills.com you get what you would expect from a simple CRM: a pipeline board with stages, deal cards you can drag between them, a journal of notes and calls on each deal, and reminders that turn amber or red when something sits too long. You can start from an industry template (real estate, recruiting, agency, home improvement, construction trades, B2B sales, nonprofit or e-commerce) instead of a blank board.

Nothing about this side requires AI. A teammate who never opens a chatbot can use it all day.

The agent side: Claude or ChatGPT through MCP

MCP (Model Context Protocol) is an open standard that lets AI apps call outside tools. The KissMySkills MCP connector, the same one that loads KissMySkills skills into your chat, also gives the agent a set of CRM tools. With them, the agent can:

• Create, update, search and open deals, and move them between stages.
• Add a note or a logged call to a deal's journal.
• Pull a "today" view: overdue steps, red and amber reminders, upcoming dated items and leads waiting to be accepted.
• Run pipeline analytics.

That is the record layer. The part that makes this a CRM with an AI agent rather than a CRM with an AI sidebar is the next one.

The agent changes the CRM itself

The same connector lets the agent change the setup, not just the data:

• Add, rename or remove pipeline stages.
• Add, update or remove custom fields.
• Set the list of lost reasons.
• Set reminder timers: days since a deal entered its stage, days since its last update, or days since a date field, each with an amber and a red threshold.
• List and apply industry templates.

Every one of these setup changes is saved as a new version. The agent can show you the history (version number, when, who and what changed) and roll back to an earlier version. Version 0 is the original default setup. A rollback creates a new version instead of deleting anything, and destructive changes such as removing a stage or field, or rolling back, need your confirmation first. After each change the agent gives you a link that opens the result in the CRM, so you can check it on the board right away.

A worked example: one round trip

Here is how this plays out for a small web agency. Emma runs client projects with two developers. Her board uses the agency template.

Morning, on the board. Emma opens the CRM in her browser and drags the "Harper Bakery redesign" card from In progress to Client review. She has sent the staging link and wants feedback.

Midday, in chat. In Claude she asks: "What is stuck in my pipeline?" The agent pulls the today view and replies that two projects have red reminders, and that Harper Bakery just moved to Client review. It already sees her drag from the morning. Nobody had to sync anything.

The setup change. Emma realizes that "Client review" hides two very different situations: waiting for feedback, and waiting for payment. She types: "Add a stage called Awaiting payment after Client review, add a date field called Invoice sent, and set a reminder that goes amber 7 days after Invoice sent and red after 14." The agent makes the three changes and returns a link. Emma clicks it and the new column and field are on her board.

Something goes wrong. Later, one of the developers asks the agent to tidy up and it renames a stage in a way that confuses everyone. Emma asks: "Show me the setup history." She sees each version with who changed what and when. She says: "Roll back to the version before the rename." The agent describes exactly what will change, Emma confirms, and the rollback is saved as a new version. The earlier history stays intact, so nothing is lost.

The same loop works for an IT team running a tickets board. A developer drags a ticket to Blocked, a lead asks the agent "which tickets have not been updated in five days?", the agent adds a "Waiting on vendor" stage, and the whole team sees it on the board.

A third use: the CRM as your agent's workspace

There is one more way to use the agent side. Instead of tracking your deals, the agent can track its own work: tasks as cards, progress as notes, status as stages. A fresh chat tomorrow reads the board and picks up where it stopped, and you watch progress in the browser. We cover that setup in detail in a CRM for your AI agent.

Who a two-way AI CRM fits

Sales teams

Reps update deals on the board between calls and ask the agent for the day's priorities in chat. Managers can ask for analytics or a stage change without filing a request with whoever "owns the CRM settings."

IT and dev teams

A pipeline works fine as a tickets board: New, Triaged, In progress, Blocked, Done. Developers who already live in Claude can log notes and move tickets from chat, while others use the board.

Freelancers and agencies

Client projects move through clear phases, and the phases change as your offer changes. Having the agent adjust stages and fields as you go, with rollback as a safety net, saves a lot of settings-screen time.

Solo founders and AI agent operators

If you are one person wearing every hat, chat is often faster than any interface. If you run AI agents on longer jobs, the board gives you a shared, visible place to see what they did. Small teams comparing options may also find our HubSpot alternative for small teams guide useful.

Setup history: a library of how your process evolved

Most CRMs keep a log of record changes but treat the setup as a single current state. Here, the setup has its own versioned history. Over months that becomes a readable story of your process: when you added a qualification stage, when you tightened a reminder from 14 days to 7, when you switched to a different template and why.

That has practical value:

• Safe experiments. Try a new stage layout for a month. If it does not help, roll back.
• Accountability. Each change records who made it, a person or an agent acting for them.
• Onboarding. A new teammate can ask the agent "how has our pipeline changed this year?" and get an answer from the history instead of from memory.

Getting started in 3 steps

1. Create your CRM. Go to crm.kissmyskills.com, sign up for free and pick an industry template or start with the default pipeline.
2. Connect your AI app. Add the KissMySkills MCP connector in Claude or ChatGPT (any MCP-capable client works). If you already use it for KissMySkills skills, the CRM tools are part of the same connector.
3. Try a round trip. Drag a card on the board, then ask the agent "what changed in my pipeline today?" Next, ask it to add a stage and open the link it returns.

Honest limits

This is a focused CRM, not a suite. To be clear about what it does not do:

• No email sync, no phone dialer and no third-party integrations.
• No mobile app. The browser version is what you get.
• No automation engine beyond what is listed here: reminder timers, templates and what the agent does when you ask.
• The agent acts when you talk to it. It does not run on its own in the background.
• The agent can make mistakes like any AI. That is why setup changes are versioned and destructive ones need your confirmation.

If you need a full sales platform with calling and email sequences, a bigger tool is the better fit. If you want a clear pipeline that you and your agent can both run and reshape, this is built for that.

Pricing

See KMS All-Access for details.

FAQ

What is an MCP CRM?

An MCP CRM is a CRM that exposes its actions through the Model Context Protocol, so an AI agent like Claude or ChatGPT can create deals, move stages and change settings directly, instead of only giving advice about them.

Is this a Claude CRM or a ChatGPT CRM?

Both. The connector works in Claude and ChatGPT, and in any other MCP-capable client. The data lives in the CRM, not in the chat, so you can switch between apps.

Can the agent break my setup?

It can make a change you do not like, but it cannot erase your history. Every setup change is a new version, destructive changes need your confirmation, and you can roll back to any earlier version, including version 0, the original default.

Do I have to use the AI at all?

No. The browser CRM works on its own. Some teammates can use only the board while others use chat. Both see the same data.

Is it only for salespeople?

No. The pipeline fits anything that moves through stages: client projects, support or dev tickets, recruiting candidates, donors, orders, or your AI agent's own task list.