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

# OCR

> Pull text out of images, scans, and photos straight into your agent.

# OCR

> **In one line:** Let your UnleashX agent read text from images, screenshots, business cards, and scanned documents.

|                    |                                                                                   |
| ------------------ | --------------------------------------------------------------------------------- |
| **Category**       | AI & Media                                                                        |
| **Authentication** | Platform-managed                                                                  |
| **Setup time**     | \~1 minute                                                                        |
| **Difficulty**     | Easy                                                                              |
| **Best for**       | Reading documents, business cards, receipts, and screenshots inside an agent flow |

## 1. Overview

OCR (Optical Character Recognition) turns pictures of text into actual, usable text. Point it at a photo of a receipt, a scanned contract, or a screenshot and it returns the words it finds — preserving structure, line breaks, and tables as much as possible.

Once connected, your agent can read text from a URL, a local file, or a base64 image, batch-process a whole folder of images, and even return word-level positions and confidence scores when you need structured output. It runs OpenAI Vision under the hood, so it is multilingual by default.

Connecting OCR to UnleashX means your voice and automation agents can act on what they "see" — extracting a phone number from a business card, reading an invoice total, or capturing a form field — without any manual data entry.

## 2. What you'll need

<Note>
  OCR is **built into UnleashX**. There is no third-party account to create and no API key for you to manage.
</Note>

* An active UnleashX account.
* The OCR feature enabled on your workspace/plan.
* Permission to edit the agent you want to add OCR to (admin or editor role). If you don't have that access, ask a workspace admin to enable it for you.

## 3. Get your credentials

<Info>
  There are **no credentials to create**. OCR is a platform-managed integration — UnleashX provisions and rotates the underlying OpenAI Vision keys for you. You never see or handle an API key.
</Info>

If you are a workspace admin, the only optional platform-side configuration is the vision model used for extraction (UnleashX sets a sensible default, e.g. a GPT-4o-class vision model). Most teams never need to change this.

| Platform-managed setting | Plain-English reason it exists                                               |
| ------------------------ | ---------------------------------------------------------------------------- |
| Vision model             | Controls quality/speed of text extraction. UnleashX picks a good default.    |
| API key                  | Provisioned and rotated by UnleashX so every agent can read images securely. |

## 4. Connect on UnleashX

<Steps>
  <Step title="Open your agent">
    Go to [https://www.tryunleashx.com](https://www.tryunleashx.com) and open the agent you want to give OCR to.
  </Step>

  <Step title="Open Data Connectors">
    In the agent, click **Data Connectors**.
  </Step>

  <Step title="Find OCR and add it">
    Locate **OCR** in the list and click **Connect** / **Add**. Because it's platform-managed, there is no key to paste — it activates immediately.
  </Step>

  <Step title="Confirm it's connected">
    The OCR tool should now show a **Connected** badge. Your agent can start reading images right away.
  </Step>
</Steps>

### Use OCR in a Workflow

Once connected, you can add **OCR** to any automation from the **Workflows** builder. Its triggers and tools appear in the Apps panel, marked with an **MCP** badge.

<Steps>
  <Step title="Add a trigger or action node">
    Open **Workflows → New Workflow**. On the canvas, click **+ Add Trigger** (or the **+** below any node) to add a step.

    <img src="https://mintcdn.com/unleashx/LXZeho1M8XzFkz0k/images/mcp/workflows/05-add-trigger.png?fit=max&auto=format&n=LXZeho1M8XzFkz0k&q=85&s=449a0c03b0c7b40e571ec02373a85259" alt="Add a trigger on the workflow canvas" width="1390" height="722" data-path="images/mcp/workflows/05-add-trigger.png" />
  </Step>

  <Step title="Pick OCR from the Apps panel">
    In the **Paths** panel, open the **Apps** tab and select **OCR**. Use the search box if you have many connectors.

    <img src="https://mintcdn.com/unleashx/LXZeho1M8XzFkz0k/images/mcp/workflows/06-apps-list.png?fit=max&auto=format&n=LXZeho1M8XzFkz0k&q=85&s=bbb8bb6e668c4f8e1f6fd5e4955a45f0" alt="Choose your app from the Apps list" width="2446" height="1550" data-path="images/mcp/workflows/06-apps-list.png" />
  </Step>

  <Step title="Choose a trigger or tool">
    Pick the trigger or action you want. Configure its fields — required fields are marked with a red asterisk (\*).
  </Step>

  <Step title="Add or select your account">
    Under **Selected account**, choose an already-connected account, or click **Add Account** to connect one now.
  </Step>

  <Step title="Save and test">
    Fill in the remaining fields and click **Save**. Use **Test** to verify the step, then toggle **Publish** when the workflow is ready.
  </Step>
</Steps>

<Note>
  The steps are the same for every connector. For the full workflow builder guide, see [Using MCP in Workflows](/mcp/workflows).
</Note>

## 5. Available tools

| Tool                | What it does                                                             | Changes data? |
| ------------------- | ------------------------------------------------------------------------ | ------------- |
| Extract Text        | Extract plain text from an image given by path, URL, or base64           | No            |
| From URL            | Download an image from a URL and extract its text                        | No            |
| From File           | Extract text from a local image file                                     | No            |
| Extract With Boxes  | Extract text plus bounding boxes and confidence details                  | No            |
| Batch Extract       | Extract text from every image in a directory                             | No            |
| Supported Languages | List the languages OCR can read                                          | No            |
| Get Info            | Report OCR server status and configuration                               | No            |
| Preprocess Image    | Clean up an image (resize, grayscale, threshold, blur) before extraction | No            |

<Note>
  Every OCR tool is read-only — it reads pixels and returns text. **Preprocess Image** writes a temporary working file but does not modify your original image or any stored data.
</Note>

## 6. Example usage

**"Read this business card and pull out the name, company, and phone number."**
→ Runs **From URL** (or **Extract Text**) to extract the text, then the agent parses the fields.

**"Go through every receipt image in this folder and give me the totals."**
→ Runs **Batch Extract** across the directory and the agent reads each total from the returned text.

## 7. Permissions & data access

**UnleashX can:**

* Read images you provide (by URL, file path, or base64) and return the extracted text.
* Optionally return word positions and confidence scores.
* Temporarily process images in memory or in a short-lived temp file.

**UnleashX cannot:**

* Edit, delete, or store your original images permanently.
* Access images you don't explicitly pass to a tool.
* Browse your device or cloud storage on its own.

**To disconnect:** Open the agent → **Data Connectors** → **OCR** → **Disconnect**. Access is revoked immediately.

## 8. Troubleshooting

| Problem                         | What it means                                     | How to fix it                                                                    |
| ------------------------------- | ------------------------------------------------- | -------------------------------------------------------------------------------- |
| "Could not load image from URL" | The URL isn't a direct image link or isn't public | Use a direct `.jpg`/`.png` URL that's publicly reachable                         |
| 401 / credential error          | The platform-managed vision key is unavailable    | This is on UnleashX's side — contact **[cs@unleashx.ai](mailto:cs@unleashx.ai)** |
| 403 / feature not enabled       | OCR isn't enabled on your plan                    | Ask a workspace admin or contact support to enable it                            |
| Empty or partial text           | Image is low-resolution, blurry, or rotated       | Run **Preprocess Image** first, or supply a clearer image                        |
| "Pillow is required"            | An image library is missing on the server         | Platform-side issue — contact support                                            |

For general connector issues, see [/mcp/integrations](/mcp/integrations).

## 9. Frequently asked questions

**Is my image data stored?**
No. Images are processed to extract text and are not retained as part of the integration. Any temp files created during preprocessing are cleaned up automatically.

**Do I need an OpenAI account or key?**
No. OCR is platform-managed — UnleashX provides and rotates the underlying vision keys.

**Which languages are supported?**
Many, automatically. Call **Supported Languages** for the current list (English, Spanish, French, German, Chinese, Japanese, Arabic, Hindi, and more).

**Can multiple team members use it?**
Yes. Once enabled on the workspace, anyone with access to the agent can use OCR.

## 10. References

* OpenAI Vision (text from images) overview: [https://platform.openai.com/docs/guides/vision](https://platform.openai.com/docs/guides/vision)
* UnleashX dashboard: [https://www.tryunleashx.com](https://www.tryunleashx.com)
* UnleashX integrations help: [/mcp/integrations](/mcp/integrations)