# Cradl AI MCP for AI Agents MCP > Cradl AI equips your agent to read and structure data from any document type, whether it's a complex invoice, a simple ID scan, or a custom form. It uses deep learning models to pull out key details—like dates, amounts, names, and IDs—and turn them into clean, usable text for your workflow. ## Overview - **Category:** artificial-intelligence - **Price:** Free - **Tags:** ocr, data-extraction, deep-learning, document-processing, automation, structured-data ## Description Dealing with documents is messy work. You get PDFs, scans, JPEGs, and forms that look different every time, making manual data entry a nightmare. Cradl AI changes that by letting your agent send the document URL directly to the system. It uses deep learning models to analyze the file and pull out exactly what you need—like invoice numbers or customer names—into structured, actionable fields. It's built specifically for high-volume data processing in finance and operations. You don't write complex parsers; you just point your agent at the document, and it does the heavy lifting. If you already use Vinkius to connect various services, adding Cradl AI keeps all your document intelligence centralized right where your agent works. After extraction, you can track everything—from listing available models to checking if a large batch of documents finished processing. It turns unstructured paper trails into clean data points that your application can use immediately. ## Tools ### extract_data_from_url Triggers a new prediction, using the OCR engine and deep learning models to pull structured data from any provided document URL. ### get_batch_details Retrieves detailed summaries and status reports for an entire group of processed documents. ### get_flow_details Shows the specific structure and configuration settings for a designated document processing workflow. ### get_model_details Retrieves detailed metadata, accuracy metrics, and schema definitions for any trained extraction model. ### get_task_status Checks the current status of a document task and resolves confidence scores alongside extracted key-value pairs. ### list_batches Provides a list of all processed document batches, including creation dates and total counts. ### list_workflows Lists every defined document processing flow, showing their associated triggers and steps. ### list_extraction_models Retrieves a comprehensive list of all data extraction models available in the system, noting their status and version. ### list_processing_tasks Lists recent document processing tasks, showing IDs, statuses (e.g., FAILED), and timestamps. ### search_models_by_name Searches for specific extraction models using a keyword query against the model metadata. ## Prompt Examples **Prompt:** ``` I just got a pile of vendor invoices. Can you extract all the invoice numbers and total amounts for me? ``` **Response:** ``` **✅ Extraction Complete.** We found 12 documents, and here is the structured summary: | Invoice ID | Vendor Name | Total Amount | Date Found | | :---: | :---: | :---: | :---: | | INV-9032A | Acme Corp | $452.11 | 2024-06-12 | | REC-77B | Beta Services | $89.00 | 2024-06-12 | | INV-1221X | Global Supply | $1,230.55 | 2024-06-11 | ``` **Prompt:** ``` The processing job for the HR documents seems stuck. What's its current status? ``` **Response:** ``` **Status Check: 🟡 PROCESSING** The task is currently running and has processed 85% of the required fields. The system predicts a completion time in about 5 minutes. * **Total Documents:** 20 * **Completed:** 16 (98% confidence) * **Failed:** 4 (Check workflow settings for root cause) Let me know if you want me to check the flow details next. ``` **Prompt:** ``` I need to find out what kinds of custom models are available for my account. ``` **Response:** ``` **🔎 Models Found:** You currently have access to 4 specialized extraction models: * **Invoice Parser v3.1:** Best for multi-page billing statements. * **ID Document Scanner:** Optimized for US/EU government IDs (high accuracy). * **Receipt Extractor v2.0:** Ideal for quick, low-detail vendor receipts. * **Custom Form Auditor:** Trained on your specific internal forms. Which one should I use next? ``` ## Capabilities ### Extract Data from URLs Pulls structured key-value pairs directly from the content of any document hosted online. ### Check Task Status and Results Confirms if a specific document processing job is finished, and provides extracted fields along with confidence scores. ### Manage Document Processing Batches Lists all processed document batches or retrieves detailed summaries for an entire group of files. ### List Available Extraction Models Shows the names, versions, and training status of every custom-trained data extraction model you own. ### Review Workflow Settings Retrieves the structure and configuration for specific document processing flows. ## Use Cases ### Processing Quarterly Expense Reports An operations manager receives 50 vendor expense reports. Instead of opening each PDF to copy dates and amounts, the agent uses the MCP to send all URLs in a batch. The system extracts every required field, giving the manager one clean spreadsheet ready for accounting. ### Onboarding New Employees A HR specialist needs to process multiple employee IDs and contracts. They feed the document links into the agent. Using specialized models, the system extracts names, dates of birth, and ID numbers accurately, allowing the onboarding workflow to continue without delays. ### Reconciling Supplier Invoices A finance analyst receives a mix of invoice formats from different suppliers. The agent uses Cradl AI’s extraction tools to read every unique document type, normalizing the data structure so it can be automatically uploaded into the accounting platform. ### Auditing Historical Documents A compliance officer needs to verify a year's worth of operational documents. They use the MCP to list workflows and check batch details, ensuring every document in the archive passed through the correct processing steps. ## Benefits - Instead of manually reading PDFs, you send the document link once. The system handles the deep learning analysis and returns clean data fields instantly. - You can monitor high-volume operations using tools like `list_batches` or `get_task_status`, giving immediate visibility into processing success rates. - Developers gain confidence by running checks on model performance via `get_model_details` before deploying the MCP to production workflows. - The ability to list available extraction models means you always know what parsing capabilities your agent has access to, minimizing integration guesswork. - When a process fails, tools like `list_processing_tasks` quickly locate the failed task ID and status, cutting down on debugging time. ## How It Works The bottom line is: you send it a document link, and you get structured data back without manual cleanup or coding boilerplate. 1. First, your agent sends a document URL to this MCP. The system runs an OCR engine that processes the image or text. 2. Next, the deep learning model attempts to predict and normalize data boundaries based on the file type (e.g., recognizing an invoice number vs. a total amount). 3. Finally, your agent receives clean JSON output containing all the extracted key-value pairs, ready for immediate use. ## Frequently Asked Questions **How does Cradl AI help me read scanned receipts and invoices?** It uses deep learning models to process images, not just digital text. You send the document link, and it performs OCR (Optical Character Recognition) to pull out key details like amounts, dates, and vendor names from poor-quality scans or photos. **Can I use Cradl AI for multiple types of documents? For example, IDs and invoices?** Yes. You can train and list multiple specialized models (like an ID scanner model separate from an invoice model). This lets your agent select the right tool for every different document type you encounter. **If I process a large batch of documents, how do I know which ones failed?** You can use the MCP to list all processed batches and then check individual task statuses. This gives you precise feedback on exactly which document caused an error and why. **Is Cradl AI just for finance, or can it handle other types of forms?** It's not limited to finance. While invoices are a core strength, the system is designed for custom data extraction. You can train models on virtually any structured document—HR onboarding forms, legal contracts, etc. **What should I do if my current model isn't working well on new documents?** Check the model details using the MCP to review its accuracy metrics. If performance is dipping, you can use the platform's features to audit and improve that specific extraction model.