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

# Connect Bento to an AI agent

> Use Bento MCP with ChatGPT, Claude, Cursor, Claude Code, Codex, and compatible clients.

Bento exposes one remote MCP endpoint:

```text theme={null}
https://mcp.bento.surf/mcp
```

It uses your existing Bento account through OAuth. You do not need to create or paste a Bento API key.

## Connect a client

> [tab] ChatGPT
>
> Open **Settings → Apps & Connectors → Advanced settings**, enable Developer mode, create a
> connector, and paste the MCP URL.

> [tab] Claude
>
> Open **Settings → Connectors**, add a custom connector named Bento, and paste the MCP URL.

> [tab] Claude Code
>
> Run:
>
> ```sh theme={null}
> claude mcp add --transport http bento https://mcp.bento.surf/mcp
> ```

> [tab] Cursor or Codex
>
> Add a remote Streamable HTTP MCP server with this configuration:
>
> ```json theme={null}
> {
>   "mcpServers": {
>     "bento": {
>       "url": "https://mcp.bento.surf/mcp"
>     }
>   }
> }
> ```

Your client opens Bento in the browser. Sign in, review the requested access, and choose **Allow access**.

## Run free tools through MCP

Ask your agent to call `list_free_tools` to see all 59 tools and their required inputs. The agent can then run text and link utilities, AI writing tools, public media URL helpers, image tools, and audio transcription through `run_free_text_tool`, `run_free_ai_tool`, `run_free_media_tool`, `run_free_image_tool`, or `transcribe_free_audio`. These runners also appear on the ChatGPT MCP endpoint.

Name the Bento workspace when you want a result associated with a particular brand. The agent will use `list_workspaces` and pass its ID. Image or audio inputs can be inline base64 or a Bento media key owned by that workspace. Image results are returned to the agent as images and are not stored automatically. Public media tools need your confirmation that you own the media or have permission to use it; their download URLs can expire. AI, media, image, and transcription limits still apply.

## What an agent can do

An authorized agent can work with pages and blocks, Store products and orders, Calendar, Community, Newsletter, social publishing, Auto DMs, analytics, referrals, and creator settings. The same plan, ownership, provider-readiness, storage, and publishing rules used by the Bento interface still apply.

Destructive or externally visible operations are marked for client confirmation. Review every publish, send, delete, moderation, or payout action before approving it.

## Disconnect

Remove or disable Bento in your client's MCP settings. Reconnecting starts a new OAuth authorization flow.

## Upload videos and create one multi-platform draft

Ask your agent: **“Upload this original video and save one draft for X,
Instagram and YouTube, with a different caption for each.”** Bento stores the
video once and creates one draft with all selected accounts. You can review and
edit each platform's caption in Post Scheduler before publishing.

For large local files, the agent should call `begin_media_upload`, upload the
original bytes in 8 MiB chunks using its file/HTTP tools, and complete the upload.
The private upload credential expires after two hours and only permits that one
upload. Current hosted paid plans accept videos up to 250 MiB per file, Free up
to 10 MiB, with sufficient remaining storage. Publishing platforms may impose
additional format, duration or size rules.

Remote MCP supports `upload_media` with inline base64 up to 25 MiB. Browser
WebMCP uses a smaller request transport, so its inline limit is 128 KiB; use
`begin_media_upload` for larger local files. Public media URLs can be streamed. Agents should not compress a video simply to fit a
chat tool's message limit. Some clients cannot access attachments or send binary
HTTP requests. In that case, upload the file using Bento's attachment picker and
ask the agent to reuse that asset.

The server supplies instructions during MCP initialization, in tool descriptions,
and through `skill://bento/bento/SKILL.md` and
`skill://bento/bento/references/tools.md`. The reference includes the exact upload
protocol and an example using `connectionOverrides[connectionId].bodyOverride`.
Clients that cache tools may need to refresh or reconnect Bento to see new tools.

## Give connected agents your Bento context

Connected agents can fetch your **Brain, Routines, Analytics and Discover** data,
including your writing voice, original source text, publishing schedule, past
performance, saved ideas and current stored trends.

Try: **“Read my Brain and recent performance, check my routines and Discover,
then draft a LinkedIn post in my voice based on a current topic.”**

The overview tools are `get_content_brain`, `get_content_routines`,
`get_analytics_workspace` and `get_content_discover`. The `read_workspace_data`
tool lets an agent page through older records and retrieve full documents by ID.
The server's bundled instructions explain every collection and pagination field.

These reads use data already stored in your workspace. They do not trigger fresh
scraping or AI generation. Your plan's history limits still apply, and an agent
cannot read another workspace or your integration credentials. Refresh/reconnect
your MCP client if it caches an older tool list.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.