MCP

MCP (Model Context Protocol) is the standard by which an AI agent discovers and uses external tools. Aikount exposes its accounting as MCP tools, so a compatible agent can issue invoices, capture expenses, reconcile the bank or query taxes without you writing any code.

There are two ways to connect, with the same accounting tools underneath. Choose based on where your agent runs.

Remote server (hosted)#

Aikount hosts an MCP server:

https://api.aikount.com/mcp

• Streamable-HTTP transport.
• Header authentication: send your token the same way as in the API, with Authorization: Bearer <token> (use an agl_ key; see Authentication).
• Nothing to install: point your MCP client at the URL and give it the token.

This is the recommended option when your agent lives in the cloud or you want to avoid local installs.

Local server (aikount-mcp)#

For agents that run on your machine, we publish the aikount-mcp package on PyPI. The quickest path is to launch it with uvx, with no permanent install:

AIKOUNT_TOKEN=agl_your_key_here uvx aikount-mcp

• stdio transport (standard input/output), the usual one for desktop MCP clients.
• Authentication via the AIKOUNT_TOKEN environment variable, holding your agl_ key.

This is the right option when your MCP client talks over stdio to local processes (many editors and agent CLIs do).

Remote or local#

Both paths operate on your company with whatever token you give them. The same cautions as an agl_ key apply here: it is full access to your accounting.

Next step#

To plug this into Claude Code, Cursor, Gemini CLI or Codex, follow Connect an agent.