Agentic Document Parsing and Extraction (ADP)

About ADP MCP Server

ADP MCP Server connects MCP-compatible AI clients to Laiye's Agentic Document Processing (ADP) platform for document parsing, extraction and workflow applications. This Node.js package runs locally over stdio, letting clients discover tools, process documents and query results from the conversation.

ADP integrates Vision-Language Models (VLM), Large Language Models (LLM) and autonomous agent decision-making technologies. It reshapes traditional document processing by transforming conventional rule-based field extraction into goal-driven end-to-end intelligent automation. Focused on intelligent processing of business documents, ADP automatically classifies overseas invoices, domestic vouchers, procurement contracts, logistics documents, financial statements and trading contracts, and accurately extracts key fields. It also supports table parsing, content verification and multilingual recognition. Requiring no template setup, data labeling or ongoing rule maintenance, it efficiently handles high-volume document processing tasks.

---

Quick Start

1. Get Your API Key

Sign up at https://adp-global.laiye.com/ and retrieve your API Key from My MCP in personal settings.

2. Configure Your MCP Client

The ADP MCP Server runs over stdio transport: the client launches a local process via npx and passes your API Key through an environment variable. Add the following to your client's MCP config file (e.g. Claude Desktop's claude_desktop_config.json, Cursor's mcp.json):

{
  "mcpServers": {
    "adp": {
      "command": "npx",
      "args": ["-y", "@laiye-adp/mcp"],
      "env": {
        "ADP_API_KEY": "<YOUR-ADP-API-Key>",
        "API_SOURCE": "npm-mcp"
      }
    }
  }
}

The same parameters apply to any MCP client:

Parameter	Value
Command	npx
Args	["-y", "@laiye-adp/mcp"]
Transport	stdio
Authentication	Environment variable ADP_API_KEY=<YOUR-ADP-API-Key>
Source marker	Optional environment variable API_SOURCE, default npm-mcp

The optional environment variable ADP_ACCEPT_LANGUAGE (zh / en, default zh) switches the API language and domain. The optional environment variable API_SOURCE sets the OpenAPI request header X-Api-Source; the default is npm-mcp. Requires Node.js 20+ installed locally.

3. Start Using

Once connected, your AI client will automatically discover all available ADP tools. Just describe what you need in natural language:

• "Upload this local file and extract fields from the returned URL"
• "Parse the structure of this PDF"
• "Extract the information from this ID card"
• "Pull the amount and date from this invoice"

---

Tool Catalog

Document Parsing

Tool Name	Title	Description
upload_temporary_file	Upload Temporary File	Upload chunk and return download_url. Pass data.download_url to parsing or extraction tools as the file parameter.
parse_document	General Document Parsing	Parse PDF, image, Word, Excel, PPT and other documents for layout analysis, returning structured text blocks, tables, reading order and page coordinates. Use when the document type is unknown and you need raw structure for further processing; for invoices, ID cards or other specific types, prefer the dedicated extraction tool.

Invoice & Order Extraction

Tool Name	Title	Description
extract_china_invoice	China Invoice / Bill	Covers 30+ common Chinese bill types: fully-digitalized e-invoices, general VAT invoices, special VAT invoices, taxi receipts, train tickets, air itineraries, fiscal invoices, etc. Extracts invoice number, date, amount, buyer, seller and other key fields; also supports invoice authenticity verification.
extract_global_invoice	Global Invoice / Receipt	Extract key fields (invoice number, date, amount, tax, currency, line items, etc.) from overseas invoices, receipts or bills in PDF or image format. Ideal for cross-border trade, reimbursement and accounts-payable automation; for Chinese VAT invoices, use the dedicated China invoice tool.
extract_purchase_order	Purchase / Sales Order	Extract key fields (order number, buyer/seller info, order date, item details, quantity, unit price, total amount, shipping address, etc.) from purchase or sales orders in PDF or image format. Suitable for e-commerce order entry, supply-chain reconciliation and warehouse management automation.

Card & Certificate Extraction

Tool Name	Title	Description
extract_id_card	China ID Card	Extract key fields (name, gender, ethnicity, date of birth, ID number, address, issuing authority, validity period, etc.) from Chinese mainland resident ID card images. Supports both front and back sides. For HK/Macao/Taiwan permits or passports, use the corresponding tool.
extract_bank_card	Bank Card	Extract key fields (card number, issuing bank, card type, expiration date, etc.) from the front side of a bank card image. Only public card-face information is recognized; sensitive fields such as CVV are not extracted.
extract_vehicle_cert	Vehicle Certificate	Extract key fields (certificate number, vehicle brand, model, VIN, engine number, manufacturing date, etc.) from vehicle certificate of conformity images. Suitable for vehicle registration, used-car transactions and fleet management.
extract_account_permit	Bank Account Permit	Extract key fields (company name, basic account number, bank name, approval number, issue date, etc.) from corporate bank-account opening permit images.
extract_driver_license	China Driver License	Extract key fields (name, gender, nationality, date of birth, license number, permitted vehicle class, first issue date, validity period, etc.) from Chinese motor-vehicle driver license images. Supports both main and supplementary pages.
extract_business_license	Business License	Extract key fields (company name, unified social credit code, legal representative, registered capital, establishment date, business scope, registered address, etc.) from Chinese business license images.
extract_passport_cn	China Passport	Extract key fields (Chinese name, romanized name, gender, date of birth, passport number, nationality, issue date, validity period, issuing authority, etc.) from PRC passport images. Foreign passports are not supported.
extract_vehicle_license	Vehicle License	Extract key fields (plate number, vehicle type, owner, VIN, engine number, registration date, issue date, etc.) from Chinese motor-vehicle license images. Supports both main and supplementary pages.
extract_org_code_cert	Organization Code Certificate	Extract key fields (organization name, organization code, legal representative, address, issue date, validity period, etc.) from organization code certificate images. For newly registered enterprises, use the Business License tool instead.
extract_household_book	Household Register	Extract key fields (household number, household type, address, family member list including name, ID number and relationship to head of household, etc.) from Chinese household register (Hukou) images. Supports both index and individual pages.
extract_hk_macao_permit	HK/Macao Travel Permit	Extract key fields (name, gender, date of birth, permit number, issue date, validity period, issuing authority, etc.) from Exit-Entry Permit for Travelling to and from Hong Kong and Macao images.

Custom Extraction

In addition to the out-of-the-box tools above, MCP provides two fixed tools for working with custom extraction apps you create on the ADP platform:

Tool Name	Title	Description
list_custom_extract_apps	List Custom Extraction Apps	List all custom document extraction apps created by the current user, returning each app's ID, name, description, labels and output field definitions. Use this to find the app_id needed by execute_custom_extract_app.
execute_custom_extract_app	Execute Custom Extraction App	Process a file using a specified custom document extraction app. Use list_custom_extract_apps first to obtain the app_id.

Workflow: Call list_custom_extract_apps to browse available apps and get the target app_id, then call execute_custom_extract_app with the app_id and file to perform extraction.

---

Workflow Applications

Tool Name	Description
list_workflow_templates	Query official templates by name with search, offset and size. Catalog IDs are for discovery.
create_workflow_application_from_template	Create and publish an account application, or reuse the corresponding runnable application.
upload_temporary_file	For workflow files, pass the created application_id and sharing_scope="application".
execute_workflow_app	Run a published account application, synchronously by default; use wait=false for an explicitly requested async run.
query_workflow_run	Query a workflow by run_id, preserving its complete output and error.

Supported template keys:

Template	template_key
Accounts payable	finance_payable
Expense reimbursement	expense_reimbursement
Financial report extraction	financial_report_extraction

Use the matching template to create and publish an application first, then upload and run using the returned account application_id. Creation does not require can_run=true in the template catalog. Repeated creation reuses an existing runnable application and preserves its template version. Do not run using the catalog's application_id.

Call create_workflow_application_from_template:

{"template_key":"financial_report_extraction"}

Successful data contains template_key, template_version, application_id, application_name, release_version_id, application_status="published" and can_run=true. Save the returned IDs. Then call upload_temporary_file:

{"chunk":"/path/to/report.pdf","application_id":"<created application_id>","sharing_scope":"application"}

Call execute_workflow_app using the upload response's download_url, file_name, content_type and optional file_size:

{
  "application_id":"<created application_id>",
  "files":[{"url":"https://files.example.com/report.pdf","name":"report.pdf","mime_type":"application/pdf"}]
}

For async runs, add "wait":false, save data.run_id, and call query_workflow_run with {"run_id":"<returned run_id>"}. A run is successful only when status="succeeded"; also inspect the output against the user's request. Do not use document task_id query tools for workflow runs. If a creation, upload or run response is uncertain, do not automatically resubmit.

Workflow changes currently require the ADP website interface. Test and republish there before running the application through the API again. These tools support creating an application from an official template; custom workflow creation and workflow editing remain website operations.

The workflow tools accept accept_language and timeout_seconds. Use the same account region for creation, upload, execution and query: zh selects the China service and en selects the global service, or configure ADP_ACCEPT_LANGUAGE consistently.

Tool Input Parameters

Document Parsing & OOTB Extraction Tools

Common input parameters for out-of-the-box tools:

Parameter	Type	Required	Description
file	string	Yes	File URL or Base64-encoded content
file_name	string	No	File name (with extension)
with_rec_result	boolean	No	Whether to include OCR intermediate results, default true
wait	boolean	No	Whether to wait synchronously for the result, default true
timeout_seconds	integer	No	Synchronous wait timeout in seconds, default 300, range 1–900

extract_china_invoice additionally supports:

Parameter	Type	Required	Description
enable_invoice_verify	boolean or null	No	Enable (true) or disable (false) verification for this extraction only. Omission or null uses the saved app configuration without changing it.

extract_global_invoice, extract_global_invoice_fast, and extract_sea_invoice_fast additionally support:

Parameter	Type	Required	Description
enable_multi_ticket	boolean	No	Split multiple invoices in one file into separate results, default true. Applies only to international and Southeast Asia fast modes without changing app configuration.
enable_field_value_translation	boolean or null	No	Enable (true) or disable (false) field-value translation for this extraction only. Omission or null uses the saved app configuration without changing it. The target language follows accept_language or ADP_ACCEPT_LANGUAGE.

upload_temporary_file

Parameter	Type	Required	Description
chunk	string	Yes	File to upload. Local file path, file:// URL, or Base64-encoded content
application_id	string	No	Created account workflow application ID; required when sharing_scope="application"
sharing_scope	string	No	application for workflow files, or user for temporary document files
accept_language	string	No	Account region and response language, zh or en
timeout_seconds	integer	No	Request timeout in seconds, default 300, range 1–900

The response contains data.download_url, which can be passed directly to parsing or extraction tools as file.

execute_custom_extract_app

In addition to the parameters above:

Parameter	Type	Required	Description
app_id	string	Yes	Custom extraction app ID (obtained from list_custom_extract_apps)

list_custom_extract_apps

No input parameters required.

Parsing/extraction file input methods:

• URL: pass a link starting with http:// or https:// as file
• Base64: pass Base64-encoded file content as file (auto-detected when not a URL)
• Local files: call upload_temporary_file first, then pass the returned data.download_url to the parsing or extraction tool.

Sync vs Async:

• wait=true (default): blocks until processing completes, returns the result directly
• wait=false: returns immediately with a task_id for later status queries

---

Tool Output

upload_temporary_file Output

{
  "code": "success",
  "message": "",
  "tips": null,
  "data": {
    "id": "ade6dcfd9b9c11f1be34d85ed35661fd",
    "file_name": "invoice.pdf",
    "file_size": 123,
    "content_type": "application/pdf",
    "download_url": "https://adp.laiye.com/web/agentic_doc_processor/laiye/file/ade6dcfd9b9c11f1be34d85ed35661fd",
    "status": "success"
  }
}

Field	Type	Description
code	string	Business status code. Success is usually success
message	string	Response message
tips	string | null	Additional message
data.id	string	File ID
data.file_name	string	File name
data.file_size	integer	File size in bytes
data.content_type	string	MIME type
data.download_url	string	Download URL. Pass this value to parsing or extraction tools as file
data.status	string	File status, returned when available

parse_document Output

{
  "task_id": "fabd7f0a4e7211f1bbc4d85ed35661fd",
  "status": 4,
  "message": "",
  "doc_recognize_result": [
    {
      "page_num": 1,
      "document_content": "Full text content of this page...",
      "document_details": [
        {
          "type": "Text",
          "text": "Paragraph content...",
          "position": [{"points": [{"x": 311, "y": 50}, {"x": 500, "y": 50}, {"x": 500, "y": 80}, {"x": 311, "y": 80}]}],
          "ocr_confidence": {
            "ocr_mean_confidence": 0.999,
            "ocr_min_confidence": 0.998,
            "is_overall_confidence": 1
          }
        },
        {
          "type": "Table",
          "text": "Column A\tColumn B\nValue 1\tValue 2",
          "position": [{"points": [...]}],
          "ocr_confidence": {...}
        },
        {
          "type": "Picture",
          "text": "https://adp.laiye.com/web/.../file/abc123",
          "position": [{"points": [...]}],
          "ocr_confidence": {...}
        }
      ]
    }
  ]
}

Field	Type	Description
task_id	string	Task ID
status	integer	Task status code
message	string	Status message
doc_recognize_result	array	Per-page recognition results
doc_recognize_result[].page_num	integer	Page number (1-indexed)
doc_recognize_result[].document_content	string	Full text of the page in reading order
doc_recognize_result[].document_details	array	Element-level details
document_details[].type	string	Element type: Text, Table, or Picture
document_details[].text	string	Text content; image URL for Picture type
document_details[].position	array	Bounding box coordinates (4 corner points)
document_details[].ocr_confidence.ocr_mean_confidence	float	Average OCR confidence (0–1)
document_details[].ocr_confidence.ocr_min_confidence	float	Minimum OCR confidence (0–1)

extract Tool Output

{
  "task_id": "91283e544e7111f18cd6d85ed35661fd",
  "status": 4,
  "message": "",
  "extraction_result": [
    {
      "field_key": "invoice_number",
      "field_name": "Invoice Number",
      "field_values": [
        {
          "field_value": "24VLT0591617",
          "field_confidence": 1.0,
          "references": []
        }
      ]
    },
    {
      "field_key": "line_items",
      "field_name": "Product Details",
      "references": [],
      "field_confidence": 1.0,
      "table_values": [
        [
          {
            "field_name": "Description",
            "field_key": "line_items_description",
            "field_values": [
              {
                "field_value": "TESLA MODEL 3",
                "field_confidence": 1.0,
                "references": "Description: TESLA MODEL 3"
              }
            ]
          }
        ]
      ]
    }
  ]
}

Regular field (no table_values):

Field	Type	Description
field_key	string	Machine-readable field identifier
field_name	string	Human-readable field name
field_values	array	Extracted values
field_values[].field_value	string	The extracted value
field_values[].field_confidence	float	Confidence score (0–1)

Table field (has table_values):

Field	Type	Description
field_key	string	Table identifier
field_name	string	Table name
table_values	array[array]	2D array: rows of cells, each cell has field_name, field_key, field_values

How to distinguish: Field object contains table_values → table field; only field_values → regular field.

Async Response (wait=false)

{
  "task_id": "fabd7f0a4e7211f1bbc4d85ed35661fd",
  "status": "running"
}

---

Task Status Codes

Status Code	MCP Status	Description
0	running	Unknown
1	running	Ready / Queued
2	running	Processing
4	success	Success
5	failed	Failed
6	failed	Cancelled

---

Supported File Formats

Format	Extensions	Notes
PDF	.pdf	Supports both scanned and electronic
Image	.jpg .jpeg .png .bmp .tiff	Supports camera photos
Word	.doc .docx	-
Excel	.xls .xlsx	-
PPT	.ppt .pptx	-

---

Authentication

ADP MCP Server uses API Key authentication, passed via the ADP_API_KEY environment variable:

"env": {
  "ADP_API_KEY": "<YOUR-ADP-API-Key>",
  "API_SOURCE": "npm-mcp"
}

• API Key is available from the My MCP page in the ADP console
• Each API Key is bound to a single user and can only access that user's apps and data
• The API Key is passed only as an environment variable to the local process and never appears in request bodies or responses
• API_SOURCE is sent to OpenAPI as the X-Api-Source request header; when omitted it defaults to npm-mcp

---

FAQ

Q: Some card/certificate tools don't appear after connecting?

A: The tool list is dynamically generated based on your initialized apps. On first connection, the system automatically initializes all out-of-the-box apps. Refresh the tool list once initialization completes to see all available tools.

Q: A card tool returns "extraction failed"?

A: Make sure the uploaded file matches the tool type (e.g., use extract_id_card for ID card images, not extract_vehicle_cert). The file must be in a supported image or PDF format.

Q: How to handle timeouts?

A: The default timeout is 300 seconds (5 minutes). You can adjust it via the timeout_seconds parameter (max 900). For large files or complex documents, use wait=false for async mode and query results later using the task_id.

Q: How does MCP compare to ADP CLI / OpenAPI?

A: All three provide equivalent functionality — the difference is in the integration method:

Method	Best For
MCP Server	AI clients (Claude Desktop, Cursor, etc.) — no code required
ADP CLI	Terminal, scripting, AI Skill integration
OpenAPI	Business system integration, backend service calls

---

License

• MCP Server: Free to connect, provided as part of the ADP platform
• ADP Service: Cloud-based AI document processing, usage-based billing

Free tier: new users receive 100 free credits per month upon registration

Get started

---

Support & Contact

• API Docs: Open API Guide
• Product Manual: Cloud Operations Manual
• Email: [email protected]
• Website: Laiye