> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usehusk.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Query Husk Finances in Natural Language with the MCP Server

> Husk MCP Server lets you talk to your finances from Claude, ChatGPT, or Gemini. Ask questions about spend, subscriptions, and invoices in plain language.

Husk MCP Server exposes your financial data through the Model Context Protocol, so you can ask questions in natural language from any MCP-compatible AI client. Instead of building custom reports or exporting spreadsheets, you can type questions like "How much did we spend on SaaS last quarter?" and get an answer backed by live Husk data.

## Connect the MCP server

<Steps>
  <Step title="Open Integrations">
    In Husk, go to **Settings** > **Integrations** and select **MCP Server**.
  </Step>

  <Step title="Generate a token">
    Click **Generate Token** and copy the MCP server URL and access key. Treat this key like a password; do not share it in public channels.
  </Step>

  <Step title="Add to your AI client">
    In Claude, ChatGPT, or another MCP-compatible client, add a new MCP server using the URL and token provided.
  </Step>

  <Step title="Test the connection">
    Ask a simple question such as "What was our total spend last month?" to confirm the client can reach Husk.
  </Step>

  <Step title="Start querying">
    Use the example prompts below to explore your data. Refine questions over time to build custom insights.
  </Step>
</Steps>

## Example prompts

* "How much did we spend on SaaS last quarter?"
* "Which subscriptions went up in price this year?"
* "Flag every overdue invoice."
* "Show me AI spend by team for the last 30 days."
* "What is our burn rate compared to last month?"
* "List all transactions without receipts from last week."

## Supported clients

<Tip>
  The Husk MCP Server works with Claude Desktop, ChatGPT with MCP plugins, and Gemini through compatible interfaces. Support for additional MCP clients is added as the ecosystem expands.
</Tip>

## Security notes

* The MCP token inherits the permissions of the Husk user who generates it. If you need read-only access for a shared client, create the token from a user with viewer permissions.
* Rotate tokens regularly from the Integrations page. Revoked tokens stop working immediately.
* Husk does not store conversation history from MCP clients.

## Related pages

* [AI Spend Overview](/ai-spend/overview) to understand how AI cost tracking fits into broader spend management
* [Multi-Bank Aggregation](/insights/multi-bank) to ensure the data you query covers all connected accounts
* [Integrations Overview](/integrations/overview) to explore Slack and accounting connections

Husk MCP Server exposes your financial data through the Model Context Protocol, so you can ask questions in natural language from any MCP-compatible AI client. Instead of building custom reports or exporting spreadsheets, you can type questions like "How much did we spend on SaaS last quarter?" and get an answer backed by live Husk data.

## Connect the MCP server

<Steps>
  <Step title="Open Integrations">
    In Husk, go to **Settings** > **Integrations** and select **MCP Server**.
  </Step>

  <Step title="Generate a token">
    Click **Generate Token** and copy the MCP server URL and access key. Treat this key like a password; do not share it in public channels.
  </Step>

  <Step title="Add to your AI client">
    In Claude, ChatGPT, or another MCP-compatible client, add a new MCP server using the URL and token provided.
  </Step>

  <Step title="Test the connection">
    Ask a simple question such as "What was our total spend last month?" to confirm the client can reach Husk.
  </Step>

  <Step title="Start querying">
    Use the example prompts below to explore your data. Refine questions over time to build custom insights.
  </Step>
</Steps>

## Example prompts

* "How much did we spend on SaaS last quarter?"
* "Which subscriptions went up in price this year?"
* "Flag every overdue invoice."
* "Show me AI spend by team for the last 30 days."
* "What is our burn rate compared to last month?"
* "List all transactions without receipts from last week."

## Supported clients

<Tip>
  The Husk MCP Server works with Claude Desktop, ChatGPT with MCP plugins, and Gemini through compatible interfaces. Support for additional MCP clients is added as the ecosystem expands.
</Tip>

## Security notes

* The MCP token inherits the permissions of the Husk user who generates it. If you need read-only access for a shared client, create the token from a user with viewer permissions.
* Rotate tokens regularly from the Integrations page. Revoked tokens stop working immediately.
* Husk does not store conversation history from MCP clients.

## Related pages

* [AI Spend Overview](/ai-spend/overview) to understand how AI cost tracking fits into broader spend management
* [Multi-Bank Aggregation](/insights/multi-bank) to ensure the data you query covers all connected accounts
* [Integrations Overview](/integrations/overview) to explore Slack and accounting connections