Introducing Fini MCP: Build, Test and Improve Your AI Support Agent from Claude, ChatGPT or Gemini

Today we’re launching Fini MCP. It puts the whole Fini platform inside the tools your team already works in: Claude, ChatGPT, Gemini, Cursor, Claude Code, Codex, even your terminal.

We’ve spent four years building AI agents that handle millions of customer conversations a month. Over that time one thing kept bothering us. Even the best agents need people working behind the scenes. Someone has to update the knowledge, connect the tools, read the conversations, fix the prompts and decide what to do next.

If an agent can handle millions of customer conversations, it should be able to help make itself better. Fini MCP is our first big step toward that.

What is Fini MCP?

Fini MCP is a Model Context Protocol server that connects your Fini workspace to AI assistants and coding agents. Once it’s connected, you can ask Claude, ChatGPT, Gemini or Cursor to build, test, analyze and improve your Fini support agent in plain English. You don’t need to open the dashboard for each task.

MCP, short for Model Context Protocol, is an open standard that lets an AI assistant use tools and data from another product. Most support MCP servers give an assistant access to tickets: it can look up a conversation or summarize a queue. Fini MCP goes further. It gives the assistant the controls of the agent itself, so the same chat that reads your conversations can change what your agent knows, test the change and send it for approval.

Which AI tools does Fini MCP work with?

Fini MCP works with any AI tool that supports MCP connectors. That includes:

• Claude, through the Fini plugin

• ChatGPT, Gemini and Cursor, as a custom MCP connector

• Claude Code, Codex and other coding agents, through the Fini skills package

• Your terminal, through the Fini skills package

It’s the same workspace and the same permissions in every tool. A draft you start in Claude shows up in the same Review queue as one started from the terminal.

With one chat, here is what you can now do.

1. Build: set up your entire AI support agent in one conversation

Give Fini MCP your help center, your past tickets, your knowledge base and your API docs. It sets up the whole Fini agent from that: knowledge, actions, test sets, tone, behavior and rules. The agent works across email, chat and voice. From creating the agent to adding a new capability later, you describe what you want.

A typical first message looks like this:

“Set up Fini for Acme Pay from our help center, our Notion and this API spec. Build the knowledge, turn the endpoints into actions, then create the agent.”

From that one message, Fini MCP:

1. Reads your sources and drafts knowledge articles, organized into folders and scoped to the right agent.

2. Turns your API endpoints into actions the agent can take mid-conversation, like looking up an order, freezing a card or handing off to a person.

3. Runs each action with sample inputs so you can see it work before a customer ever triggers it.

4. Creates the agent and builds test sets from your past conversations.

5. Sends everything to Review, so nothing goes live until someone on your team approves it.

Adding a capability later works the same way. Say “I want the agent to process refunds under $50 and hand anything larger to a person,” and Fini MCP drafts the action and the rule for you to check.

How do you test an AI customer service agent before it goes live?

With Fini MCP, you test it by replaying conversations you’ve already had. Ask “Run the last 20 refund conversations through the draft and tell me what changed.” Fini replays them against the draft knowledge and rules and grades every reply against your criteria. It uses recorded action responses, so testing never touches a live customer or system.

You get a pass or fail on each case and the exact reply that failed. Fix it, run the set again, and publish when it passes. Chatbot testing stops being a launch-week scramble and becomes a step you can repeat every time anything changes.

2. Self-improve: make your agent better every week

A self-improving AI agent learns from its own conversations and proposes its own fixes, while a person decides what ships. With Fini MCP, you ask “Look at the 10,000 conversations we handled last week. How can we make our agent 10% better?” Fini finds the biggest gaps and drafts the changes that would close them.

Fini reads the conversations and the decisions behind each reply: what the agent searched for, which rule it followed and why it escalated. It ranks the biggest gaps in resolution, accuracy and customer experience. Then it drafts the fix for each one:

• A knowledge article for a question customers keep asking that your help center doesn’t cover

• A new action for a request the agent currently escalates but could resolve

• A rule change where your policy and your agent disagree

Each draft is tested against past conversations, so you see the old answer next to the new one before you approve anything. Approve what works and the case joins your test set, so the same mistake doesn’t come back. Next Monday, ask again. The more conversations your agent handles, the more it has to learn from.

3. Reporting: ask anything about your customers

Fini MCP turns your support conversations into answers you can ask for in plain English. Ask “What were the top five reasons customers contacted us this week, and what changed from last week?” You get the answer backed by the conversations behind it, without building a report or digging through a dashboard.

Your support conversations hold a lot about your customers, your product and your business. Until now, getting at it meant navigating dashboards, building reports or reading conversations one by one. With Fini MCP, customer support analytics becomes a question:

• “Which issues grew fastest this month?”

• “Show me every conversation where a customer mentioned cancelling, and why.”

• “QA last week’s replies on billing questions and flag anything off-policy.”

• “Send me this report every Monday at 9.”

Every number links back to the conversations it came from, so you can check the answer instead of trusting a chart.

4. Monitor: tell your AI agent what to watch for

AI agent monitoring with Fini MCP starts with a sentence. Tell Fini “If more than 20 customers report the same bug in a day, let me know.” From then on it watches every conversation against that line and tells you the moment it’s crossed, with the conversations attached.

You can set a line for anything you can describe: a bug spreading, resolution dropping below a target, a shift in customer sentiment, a spike in refund requests after a release. When something crosses it, you hear about it first. You don’t find out a week later in a report.

How does Fini MCP keep changes safe?

Fini MCP reads freely and never changes your live agent on its own. Knowledge drafts land in Review. Prompt, rule and tag changes stop for your approval before they apply. The same rules hold in Claude, ChatGPT, Gemini, Cursor and the terminal, so a person always decides what reaches customers.

This matters more than it sounds. An assistant that can only read tickets can’t do much harm. One that can change what your support agent says can, which is why the permission model works the same way everything at Fini works. The AI drafts. You decide. In Claude, set Fini’s read-only tools to “Always allow” and keep write tools on “Needs approval,” so you can explore freely and approve each change on purpose.

How do you set up Fini MCP?

Setting up Fini MCP takes about two minutes. In Claude, install the Fini plugin from Settings and sign in. In other assistants that support custom MCP connectors, such as ChatGPT, Gemini or Cursor, add Fini MCP as a connector and sign in. For coding agents and your terminal, run one install command. Chat assistants need no API key; coding agents ask once for your Fini workspace key.

In Claude (Pro, Max, Team or Enterprise plan):

1. Open Settings, then Plugins. Choose Add, then Add marketplace, then Add from a repository.

2. Paste https://github.com/ask-fini/fini-skills and keep Sync automatically on.

3. In the fini-skills marketplace, click Install on Fini.

4. In the plugin's Connectors tab, click Install on the fini row, click Add in the prefilled dialog and sign in to your Fini workspace.

If you added a custom Fini connector before, remove it first.

In ChatGPT, Gemini, Cursor and other assistants that support custom MCP connectors, add this URL as a custom connector and sign in:

https://mcp.usefini.com/mcp

In Claude Code, Codex or your terminal, run this in your project root. On first use, the agent asks for your Fini workspace key (Deploy, then API Keys):

npx skills add ask-fini/fini-skills

Then try your first question: “List my Fini agents and show the knowledge folders for each one.”

Full setup guides: Fini in Claude and Fini in coding agents.

If you’re not on Fini yet, book a call and we’ll show you what this looks like with your own customer conversations.

We’ve spent years making AI agents better at supporting customers. Now we’re giving them the ability to help make themselves better.

FAQs

1. What is Fini MCP?
Fini MCP is an MCP server that connects your Fini workspace to AI tools like Claude, ChatGPT, Gemini, Cursor, Claude Code and Codex. It lets you build, test, analyze and improve your AI customer support agent by asking in plain English, instead of doing each task in the Fini dashboard.

2. What is an MCP server?
An MCP server is a service that exposes a product’s tools and data to AI assistants through the Model Context Protocol, an open standard. Once an assistant connects to it, the assistant can take actions in that product on your behalf, within the permissions you grant, such as reading records or drafting changes.

3. Which AI tools work with Fini MCP?
Fini MCP works with Claude, ChatGPT, Gemini, Cursor, Claude Code, Codex and any other assistant that supports custom MCP connectors. It also runs from the terminal. Every tool connects to the same Fini workspace, with the same permissions and the same Review queue.

4. How do I add Fini MCP to Claude?
In Claude (Pro, Max, Team or Enterprise plan), open Settings, then Plugins, choose Add marketplace and add the ask-fini/fini-skills repository from GitHub. Install the Fini plugin, then in its Connectors tab click Install on the fini row, confirm and sign in to your Fini workspace. The plugin needs no client ID, secret or API key.

5. How do I connect Fini MCP to ChatGPT or Gemini?
Add https://mcp.usefini.com/mcp as a custom MCP connector in any assistant that supports one, such as ChatGPT, Gemini or Cursor, then sign in with your Fini account. The account you sign in with decides which workspace the assistant can see. Setup takes about two minutes.

6. Can Fini MCP build an AI support agent from scratch?
Yes. Give it your help center, past tickets, knowledge base and API docs, and it sets up the agent: knowledge, actions, test sets, tone, behavior and rules, across email, chat and voice. Everything arrives as a draft in Review, and nothing goes live until your team approves it.

7. Can Fini MCP change my live agent without approval?
No. Fini MCP reads freely, but every change waits for a person. Knowledge drafts land in Review, and prompt, rule and tag changes stop for your approval before they apply. The same safeguards hold in every AI tool, so nothing reaches customers without sign-off.

8. How does Fini MCP make an AI agent self-improving?
Ask it to review last week’s conversations and find ways to improve. It reads each conversation and the decision behind each reply, finds the biggest gaps, drafts the knowledge, action or rule changes that close them, and tests them on past conversations. You approve what works, then repeat weekly.

9. How is Fini MCP different from a Zendesk or Intercom MCP server?
Helpdesk MCP servers mainly let an AI assistant read and work with tickets and conversations. Fini MCP gives the assistant the controls of the support agent itself. It can build the agent, change its knowledge, actions and rules, test those changes and monitor performance, with every change routed through human approval.

10. Do I need to be a Fini customer to use Fini MCP?
Fini MCP connects to a Fini workspace, so you need a Fini account to use it. If you’re not on Fini yet, you can book a call and the team will set it up with your own customer conversations so you can see it working.