Connect Claude and Cursor to Your Data with MetricChat's MCP Server

Most teams already have AI tools embedded in their daily workflow — Claude Desktop for research and writing, Cursor or Windsurf for writing code. The problem is that these tools operate in isolation from your actual data. You copy a number from a dashboard, paste it into a prompt, and hope the AI reasons correctly from a static snapshot. That gap between your AI tools and your live data is where bad decisions get made.

MetricChat closes that gap with a built-in MCP server. Once connected, any MCP-compatible AI tool can query your databases directly, in real time, using plain English — without leaving the application you are already working in.

What Is MCP?

MCP stands for Model Context Protocol. It is an open standard that defines how AI applications can connect to external tools and data sources. Think of it as a structured way for an AI model to reach outside of its context window and interact with systems — running queries, fetching records, and reading live state — rather than reasoning solely from information you paste into a chat window.

Claude Desktop, Cursor, Windsurf, and a growing number of other AI clients support MCP natively. When you configure an MCP server in one of these tools, the AI gains the ability to call the tools that server exposes. MetricChat's MCP server exposes your connected data sources as a set of queryable tools.

What MetricChat Exposes Over MCP

MetricChat's MCP server provides six tools that cover the full analytics workflow:

Tool	What It Does
`get_context`	Returns available data sources, tables, and column metadata — with optional regex filtering
`inspect_data`	Pulls a 3-row sample from any table for quick exploration and validation
`create_report`	Opens an analysis session with your data sources attached
`create_data`	Generates charts or tables saved as tracked visualizations in MetricChat
`list_instructions`	Retrieves your team's organizational knowledge and AI guidance rules
`create_instruction`	Adds new AI guidance with auto-versioning and approval workflows

When Claude or Cursor calls these tools, they run against your actual live data through MetricChat's governed layer — including your defined instructions, access controls, and audit logging. The AI does not bypass your data governance. It works through it.

Setting Up the MCP Connection

Step 1: Enable the MCP Server

In MetricChat, navigate to Settings > Integrations and toggle the MCP Server on. Then click Regenerate Token to generate an access token. This token is what authenticates your AI clients against the server.

Keep this token secure. Any client holding it can query your connected data sources through MetricChat.

Step 2: Configure Claude Desktop

Claude Desktop reads its MCP server configuration from a JSON file on your local machine.

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the MetricChat server block to that file:

{
  "mcpServers": {
    "metricchat": {
      "url": "https://YOUR_DOMAIN/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Replace YOUR_DOMAIN with the hostname where your MetricChat instance is running, and YOUR_API_KEY with the token you generated. Restart Claude Desktop and the MetricChat tools will appear in the tool picker.

Step 3: Configure Cursor

Cursor has native MCP support built into its settings. Open Cursor Settings > MCP and add a new server entry using the same configuration:

{
  "mcpServers": {
    "metricchat": {
      "url": "https://YOUR_DOMAIN/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Once saved, Cursor's AI will have access to the MetricChat tools in any Composer or chat session where MCP tools are enabled.

The same configuration block works for Windsurf and any other client that implements the MCP standard.

What You Can Do Once Connected

With the integration active, you interact with your data through natural language in whichever tool you are already using. A few examples of what this looks like in practice:

In Claude Desktop:

What were our top 10 customers by revenue last quarter?

Compare month-over-month churn rate for the past six months, broken down by plan tier.

Which sales reps closed the most deals in January, and what was the average deal size for each?

Claude will call get_context to understand your schema, run the appropriate query via create_report or create_data, and return a result — all within the conversation.

In Cursor, while writing backend code:

How many rows are currently in the events table and what does the schema look like?

Pull a sample of recent error events so I can validate this parsing logic against real data.

Check whether the user_id column in subscriptions has any nulls.

Rather than switching to a database client, opening a new tab, and manually writing an exploratory query, you ask the question and continue writing code. The AI handles the data lookup inline.

Why This Matters for Your Workflow

The main benefit is not that the AI can query your database — MetricChat's web interface already does that. The benefit is that the AI can query your database from wherever you are working.

No context switching. Engineers debugging a data pipeline can validate assumptions in Cursor without leaving their editor. Analysts writing a report in Claude Desktop can pull live figures without tabbing over to a separate tool.

Real-time data. There are no exports, no cached snapshots, no CSVs to upload. Every query runs against your live data sources at the moment you ask.

Governed access. MetricChat applies the same instructions, table permissions, and audit logging to MCP-initiated queries that it applies to queries from the web interface. Your AI clients inherit your data governance rules automatically.

Tracked outputs. When the AI calls create_data to produce a visualization, that output is saved in MetricChat and attributed to an MCP session. You get full visibility into what was queried and what was generated, even when the work happens outside the MetricChat UI.

A Note on Privacy

MetricChat logs which MCP tools were called and what data was returned, but it does not save the prompt text you send to Claude or Cursor. Your conversations stay private to your AI client. The audit trail covers data access, not the surrounding dialogue.

Getting Started

If MetricChat is already running in your environment, enabling the MCP server takes about two minutes. Navigate to Settings > Integrations, generate a token, drop the configuration block into your client's config file, and restart the application.

If you are not running MetricChat yet, you can spin up a local instance with a single Docker command and connect your first data source before enabling MCP:

docker run --pull always -d -p 3000:3000 metricchat/metricchat

From there, the same setup steps apply.

The tooling you rely on to do your best work should have access to the data your business runs on. MetricChat's MCP server makes that connection straightforward — without requiring you to change how you work or compromise on data governance.

For full configuration details, see the MCP Server documentation.