Skip to content

mcp-integration.md

Latest commit

 

History

History
97 lines (66 loc) · 3.9 KB

mcp-integration.md

File metadata and controls

97 lines (66 loc) · 3.9 KB

DocExtract MCP Integration

DocExtract exposes document extraction and semantic search as MCP (Model Context Protocol) tools, allowing any MCP-compatible agent — Claude Desktop, custom agent frameworks — to process documents and query knowledge bases without custom integration code.

Available MCP Tools

Tool Description Required Inputs
extract_document Download a document from a URL and extract structured data using DocExtract AI. Supports PDF, images (PNG/JPEG/TIFF), email (.eml), and plain text. Returns the extracted record with document type, fields, and confidence score. file_url (string)
search_records Semantic search over all extracted records in DocExtract. Returns matching records with similarity scores. query (string)

Tool Input Schemas

extract_document

Parameter Type Required Description
file_url string Yes URL of the document to download and extract
doc_type_hint string No Optional type hint: invoice, purchase_order, receipt, bank_statement, identity_document, medical_record

search_records

Parameter Type Required Description
query string Yes Natural language search query
limit integer No Max results to return (default: 5, max: 100)

Quick Setup

1. Install dependencies

pip install mcp httpx

2. Set environment variables

export DOCEXTRACT_API_URL="http://localhost:8000/api/v1"
export DOCEXTRACT_API_KEY="your-api-key"

Use the public demo key for read-only access: demo-key-docextract-2026

3. Configure Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "docextract": {
      "command": "python",
      "args": ["/path/to/docextract/mcp_server.py"],
      "env": {
        "DOCEXTRACT_API_URL": "http://localhost:8000/api/v1",
        "DOCEXTRACT_API_KEY": "your-api-key"
      }
    }
  }
}

See docs/claude_desktop_config.example.json for a copy-pasteable template.

4. Use in Claude

Once configured, Claude can call DocExtract tools directly:

"Extract the invoice at https://example.com/invoice.pdf and tell me the total amount due."

"Search my documents for contracts mentioning 'indemnification'."

"Find all purchase orders from Acme Corp."

How It Works

DocExtract's MCP server (mcp_server.py) wraps the REST API as typed tools. The server runs as a subprocess, communicates over stdio using the MCP protocol, and provides JSON Schema tool definitions that Claude uses for automatic tool selection.

extract_document flow:

  1. Downloads the file from the provided URL
  2. Uploads it to the DocExtract API (POST /documents/upload)
  3. Polls the job status every 2 seconds until terminal state (up to 300s)
  4. Fetches and returns the extracted record as JSON

search_records flow:

  1. Sends the query to GET /records/search with the configured API key
  2. Returns matching records with similarity scores as JSON

Other MCP Hosts

The same mcp_server.py works with any MCP-compatible host. For Cursor or other editors, consult their MCP host documentation and use the same command/args/env pattern shown above.

Related Projects

  • mcp-server-toolkit — 9 pre-built MCP servers and a framework for building new ones. Published to PyPI (pip install mcp-server-toolkit==0.2.0). Provides production boilerplate for building MCP servers with caching, rate limiting, and OpenTelemetry instrumentation — the same patterns used in this server.
  • EnterpriseHub — Multi-agent orchestration platform that uses MCP for tool dispatch across 22 specialized agents.