MCP

Use MCP to connect approved AI assistants to Kick for client lookup, review, reporting, and follow-up drafting.

MCP, or Model Context Protocol, lets an AI assistant call approved Kick tools with your authorization. Instead of copying client data into a chat, the assistant can request the specific Kick data it needs and return structured results. Follow the Quickstart Guide to connect Kick to Claude or ChatGPT.

Use MCP when you want to work in natural language:

• "Which Acme transactions have been uncategorized for more than 30 days?"

• "Compare Acme's January and February P&L and flag categories that grew over 30%."

• "What changed in Acme's books in the last 7 days, and what needs a second look before close?"

• "Draft a client-friendly email for Acme's open transaction questions."

MCP is best for personal assistant workflows where an accountant stays in the loop. For repeatable scripts, exports, scheduled jobs, and agent loops, use the CLI.

Kick does not currently provide a one-click official connector in every AI client. Use the custom MCP setup supported by your AI client.

How MCP connects

Kick supports two connection patterns:

• Hosted MCP over HTTP - the assistant connects to Kick's remote MCP endpoint.

• Local stdio through the Kick CLI - the assistant runs kick mcp serve on your computer.

Use hosted MCP when your assistant supports remote HTTP MCP. Use local stdio when your assistant expects a local command or when you want the MCP server to share your CLI profile and workspace defaults.

Pages in this section

• MCP Quickstart — connect Claude or ChatGPT in under five minutes.

• Advanced MCP Setup — Cursor, Gemini CLI, Perplexity, generic clients, config files, PATs, OAuth scopes, and tool discovery.

• MCP Example Workflows — accountant-friendly prompts for review and reporting.

• MCP Troubleshooting — common setup, auth, workspace, and tool-list issues.