Alpha release. Install with pip install adx. APIs may change before v1.0. Roadmap →
Skip to content

MCP Tools

ADX provides a fully implemented MCP (Model Context Protocol) server, making all document tools available to Claude and other MCP-compatible AI agents.

Setup

Add ADX to your MCP client configuration:

json
{
  "mcpServers": {
    "adx": {
      "command": "adx",
      "args": ["mcp"],
      "env": {
        "ADX_STORAGE_DIR": "/path/to/storage"
      }
    }
  }
}

Available Tools

ToolDescriptionRequired Params
adx_uploadUpload and process a document filefile_path
adx_profileProfile a document: metadata, type, confidencefile_id
adx_structureGet document structure: sections, tables, pagesfile_id
adx_searchFull-text search within a single documentfile_id, query
adx_search_corpusSearch across all uploaded documentsquery
adx_get_pageGet text and tables from a specific pagefile_id, page_number
adx_get_tableGet a specific table by IDfile_id, table_id
adx_to_markdownExport a document as markdownfile_id
adx_chunkChunk a document for retrieval pipelinesfile_id
adx_list_sheetsList workbook sheets in a spreadsheetfile_id

Tool Details

adx_upload

Upload and process a document file. Returns file ID and metadata.

json
{
  "file_path": "/path/to/document.pdf"
}

adx_profile

Profile a document — file type, page/sheet count, detected document types, confidence scores, and recommended next tools.

json
{
  "file_id": "abc123"
}

Search text and cells within a single document.

json
{
  "file_id": "abc123",
  "query": "revenue forecast",
  "max_results": 20
}

adx_search_corpus

Search across all uploaded documents. Useful for finding information across a collection.

json
{
  "query": "total revenue",
  "max_results": 20
}

adx_chunk

Chunk a document for retrieval pipelines (RAG).

json
{
  "file_id": "abc123",
  "strategy": "section_aware",
  "max_chunk_size": 1000
}

Strategies: fixed_size, section_aware, table_only.

Python Usage

python
from adx.mcp.server import create_mcp_server

server = create_mcp_server(storage_dir="./data")

The MCP server wraps the same core functions as the REST API and Python SDK, ensuring consistent behavior across all interfaces.

Documents are not text blobs. Give your agents document tools.