# AI agents & MCP

There are two ways to work with banta from AI tools:

1. **Let the agent read these docs** so it can write integration code against the API for you. See "Docs for AI agents" below.
2. **Connect the agent directly to your organization's data** through the banta MCP server, so it can search, summarize and take actions itself. See "Connecting over MCP".

## Docs for AI agents (LLM version)

All of this documentation is also published as clean text that language models read easily, and it updates automatically with every API change. English (the default) lives at the site root; Hebrew lives under `/he/`.

| English | Hebrew | Contents |
|---|---|---|
| [`/llms.txt`](https://docs.banta.co.il/llms.txt) | [`/he/llms.txt`](https://docs.banta.co.il/he/llms.txt) | Index following [llmstxt.org](https://llmstxt.org): a link to every guide and every endpoint |
| [`/llms-full.txt`](https://docs.banta.co.il/llms-full.txt) | [`/he/llms-full.txt`](https://docs.banta.co.il/he/llms-full.txt) | The entire documentation in one file: guides and the full reference |
| `/guides/<slug>.md` | `/he/guides/<slug>.md` | One guide as Markdown, e.g. [`/guides/authentication.md`](https://docs.banta.co.il/guides/authentication.md) |
| `/reference/<operationId>.md` | `/he/reference/<operationId>.md` | One endpoint as Markdown, e.g. [`/reference/createLead.md`](https://docs.banta.co.il/reference/createLead.md) |
| [`/openapi.json`](https://docs.banta.co.il/openapi.json) · [`/openapi.yaml`](https://docs.banta.co.il/openapi.yaml) | [`/he/openapi.json`](https://docs.banta.co.il/he/openapi.json) | The full OpenAPI document, for Postman, Insomnia or SDK generators |

**On the site:**

- **Ask AI** (top bar) - opens ChatGPT, Claude or Perplexity with an instruction to start from `llms.txt` (and `llms-full.txt` for the complete docs).
- **Copy as Markdown** (on every endpoint) - copies that endpoint's documentation, ready to paste into a chat.

**For coding agents (Claude Code, Cursor, Copilot)** - add this to your prompt or to your project's instructions file:

```text
The banta API reference is at https://docs.banta.co.il/llms.txt (index) and
https://docs.banta.co.il/llms-full.txt (full). Read it before writing integration code.
Auth: header x-api-key. Base URL: https://app.banta.co.il/api-service
```

## Connecting over MCP

The banta MCP server gives AI agents tools for working with your organization's data: searching leads and files, summarizing a deal, creating tasks and reminders, and more. Server URL:

```text
https://db.banta.co.il/functions/v1/mcp-server
```

### Authentication

- **OAuth (recommended)** - paste the URL into your MCP client. The client sends you to a banta sign-in and consent screen, and everything else is automatic. By default only organization admins can approve a connection. You can open this up to all members, or turn external connections off entirely, in [Organization settings → Connections & technology → AI Settings](https://app.banta.co.il/organization/settings?group=integrations&tab=ai).
- **Access token** - for clients that support custom headers. Create a token (it starts with `bnt_mcp_`) on the same settings screen and send it as `Authorization: Bearer <token>`. Clients that reserve the `Authorization` header can send `X-MCP-Token: <token>` instead.

### Per client

- **Claude (claude.ai and the Desktop app)** - Settings → Connectors → Add custom connector, paste the server URL and approve on the banta screen.
- **Claude Code** - `claude mcp add --transport http banta https://db.banta.co.il/functions/v1/mcp-server`
- **Cursor / VS Code** - the install button in the top bar of this site adds the server in one click.
- **Other MCP clients** - any client that supports remote MCP servers over HTTP can connect with the server URL above: with OAuth if it supports it, otherwise with an access token. In ChatGPT this depends on your plan offering custom connectors.

### What an agent can do

- The available tools are set by your organization's policy, and can be narrowed further per token.
- Sensitive actions (money, status changes and deletions) may require approval by a user in the app. In that case the agent receives `approval_required`, and the action runs only after the request is approved in banta.
- Each connected client is limited to 60 requests per minute.
- Every agent call is recorded in the organization's audit log.

---

Part of the banta API documentation. Full index: https://docs.banta.co.il/llms.txt