---
title: "AI Toolkit · COUNT Partner API"
description: "Ship a COUNT Partner API integration with an AI coding assistant, or automate workspace tasks from an agent using the MCP server and COUNT CLI."
canonical: "https://developers.getcount.com/ai"
source: "https://developers.getcount.com/ai"
---
AI

# Build with COUNT's AI toolkit

Most partners start with the Prompt Generator to ship a Partner API integration with an AI coding assistant. Use MCP or the CLI when you want workspace automation inside an agent.

[Integration-first workflow Recommended Prompt Generator Recommended Most partners start here to scaffold OAuth, HMAC, and REST code in Cursor or Claude Code. Compose detailed prompts for bank feeds and REST endpoints against the Partner API. OAuth & token refresh Signed HTTP examples Bank connection flows Open Prompt Generator](https://developers.getcount.com/tools/prompt-generator)

[Local agent setup COUNT CLI OAuth login on your machine and a stdio MCP server for Claude Code, Cursor, and other agent runtimes. Local credentials MCP config export Workspace login Set up the CLI](https://developers.getcount.com/tools/count-cli)

[Workspace automation MCP Server 0 typed tools for workspace data, over stdio locally or OAuth remotely for Claude.ai connectors and ChatGPT plugins. COUNT_* tool catalog Remote OAuth Same API paths Explore MCP tools](https://developers.getcount.com/tools/mcp)

[Ready-made copy Prompt Library Browse copy-paste prompts by category and difficulty. Integration prompts include REST snippets; MCP prompts target COUNT_* tools. 17 categories Difficulty tags Code snippets Browse prompts](https://developers.getcount.com/tools/prompt-library)

## Which surface should I use?

Match the tool to your build target. You can use more than one — many teams generate integration prompts first, then wire MCP for internal ops automation.

### Your product on the Partner API

OAuth, HMAC, REST. Build COUNT into your own app or backend.

[Prompt Generator](https://developers.getcount.com/tools/prompt-generator)

### A coding agent on your laptop

Claude Code, Cursor, or Windsurf with local login and stdio MCP.

[COUNT CLI](https://developers.getcount.com/tools/count-cli)

### A hosted AI connector

Claude.ai or ChatGPT calling COUNT over remote OAuth MCP.

[Remote MCP](https://developers.getcount.com/tools/mcp)

Remote MCP endpoint

`https://api.getcount.com/mcp` [Setup guide](https://developers.getcount.com/tools/mcp)

## Give your assistant the docs

Before you install anything: if your AI tool accepts a documentation URL, point it at one of these. Both are generated from this site on every deploy, so an assistant reading them is never working from a stale copy of the reference.

[`/llms.txt` An index of every page — guides, endpoints, and tools — one line each. Small enough to hand to an assistant up front so it knows what exists before it starts guessing.](https://developers.getcount.com/llms.txt) [`/llms-full.txt` The entire documentation set inlined as markdown in a single file. One fetch instead of crawling every link — best when the tool has a large context window.](https://developers.getcount.com/llms-full.txt)

Grounding beats guessing

The Partner API's HMAC signing scheme is specific to COUNT, and a model without the docs in context will confidently invent a plausible-looking one. Giving it `llms.txt` first is the cheapest fix for wrong generated code.

## Connect an MCP client

Install the remote MCP server to let an agent read and write real workspace data. Full tool catalog and local (stdio) setup on the [MCP Server](https://developers.getcount.com/tools/mcp) page.

### VS Code

[Install in VS Code](https://vscode.dev/redirect/mcp/install?name=count&config=%7B%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fapi.getcount.com%2Fmcp%22%7D)

Or add this to `.vscode/mcp.json` in your workspace:

.vscode/mcp.json

```json
{
  "servers": {
    "count": {
      "type": "http",
      "url": "https://api.getcount.com/mcp"
    }
  }
}
```

### Cursor

[Install in Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=count&config=eyJ1cmwiOiJodHRwczovL2FwaS5nZXRjb3VudC5jb20vbWNwIn0=)

Or add this to `~/.cursor/mcp.json`:

~/.cursor/mcp.json

```json
{
  "mcpServers": {
    "count": {
      "url": "https://api.getcount.com/mcp"
    }
  }
}
```

### Claude Code

Terminal

```bash
claude mcp add --transport http count https://api.getcount.com/mcp
```

Run it in your project directory. Claude Code opens the browser for OAuth on the first tool call; check the connection with `/mcp`.

### Claude.ai

Settings → Connectors → Add custom connector, then paste the server URL.

Available on Pro, Max, Team, and Enterprise plans. Claude completes the OAuth consent flow in the browser and stores the connection on your account.

### ChatGPT

Open Plugins → Browse plugins, select the + button at the top, then choose Create app → MCP app. Paste the server URL as the MCP endpoint.

Requires a Plus, Pro, Business, Enterprise, or Edu account with developer mode enabled.

### Any other MCP client

Server URL

```text
https://api.getcount.com/mcp
```

MCP is an open protocol. Point any client that supports a streamable HTTP server with OAuth at this URL — it will register the same `COUNT_*` tools under the name `count`.

## Quick start

For local MCP in a coding agent, install the CLI, sign in, and print MCP configuration:

Terminal

```bash
npm install -g @countfinancial/cli
count login
count mcp print-config
```

Two paths, one API

**Build an integration** with the Prompt Generator (REST + OAuth + HMAC). **Automate a workspace** with CLI or remote MCP (`COUNT_*` tools). Local and remote MCP expose the same tool names and partner API paths.