ClicheFactory — Structured Document Extraction

ClicheFactory extracts structured, Pydantic-validated data from documents. Give it a file and a JSON schema — it returns clean JSON.

This skill requires the clichefactory MCP server. Register it once:

openclaw mcp set clichefactory '{"command":"uvx","args":["clichefactory-mcp"],"env":{"CLICHEFACTORY_API_KEY":"cliche-your-key"}}'

Get your API key at clichefactory.com → Settings → API Keys. Free pages included, no credit card.

Verify with openclaw mcp list.

MCP Tools

extract

Primary tool. Takes a document file path and a JSON schema, returns structured JSON matching the schema.

Parameters:

• file — absolute path to the document
• schema — JSON schema (inline object or path to a .json file)
• mode — "service" (default, ClicheFactory cloud) or "local" (BYOK, advanced)
• extraction_mode — omit for default OCR+LLM, or "fast" for multimodal LLM without OCR
• model — LLM override, e.g. "openai/gpt-4o"

to_markdown

Converts a document to readable markdown. Use this to inspect contents before building a schema, or when extract fails and you need to understand the document structure.

Parameters:

• file — absolute path to the document
• mode — "service" (default) or "local"

doctor

Checks configuration, Python dependencies, and system binaries. Call this when extraction fails to diagnose the issue.

Workflow

1. If the user provides a document without a schema, call to_markdown first to see the content.
2. Build a JSON schema from what you see (or from the user's description of what they need).
3. Call extract with the file and schema (service mode is the default).
4. If extraction returns validation errors, inspect them — the schema may need adjustment (wrong types, missing fields, etc.).
5. If extraction fails entirely, call doctor to check whether the API key is configured.

Local mode (advanced)

For BYOK / on-machine extraction, install local extras and set LLM credentials:

openclaw mcp set clichefactory '{"command":"uvx","args":["clichefactory-mcp"],"env":{"LLM_MODEL_NAME":"gemini/gemini-3-flash-preview","LLM_API_KEY":"YOUR_KEY"}}'

Requires pip install "clichefactory-mcp[local]".

Supported File Types

PDF, PNG, JPG, JPEG, WebP, GIF, BMP, DOCX, DOC, ODT, XLSX, CSV, EML, TXT, MD.

Schema Tips

Schemas follow standard JSON Schema. Example for an invoice:

{
  "type": "object",
  "properties": {
    "invoice_number": {"type": "string"},
    "date": {"type": "string", "description": "Invoice date in YYYY-MM-DD format"},
    "vendor": {"type": "string"},
    "line_items": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "description": {"type": "string"},
          "quantity": {"type": "number"},
          "unit_price": {"type": "number"},
          "total": {"type": "number"}
        }
      }
    },
    "total_amount": {"type": "number"}
  }
}

Use description fields in the schema to guide extraction — the LLM reads them.

More Information

• SDK and CLI: clichefactory on PyPI
• MCP server: clichefactory-mcp on PyPI
• Website: clichefactory.com