Toa is a library for converting documents to HTML.

Tip

By default, it returns a string-type HTML fragment. You can decide how to render it yourself, such as beautifying the page display.

The built-in HTML templates can return fully structured HTML content, supporting Mermaid stylesheets, formulas, and syntax highlighting, which can be enabled selectively.

Toa supports the following conversions:

• Ebooks: EPUB
• Emails: EML, MSG(outlook)
• Documents: DOCX, PDF, TXT, MD(Markdown)
• Spreadsheets: XLSX, XLS, CSV
• Presentations: PPTX
• Data Exchange: JSON, JSONL
• Markup Languages: XML, HTML
• Multimedia: audio files, video files, supports audio-to-text conversion and video subtitles extraction.
• Resources from the Internet
• ...

Toa provides convenience for online previewing of documents, emails, data reports, e-books, etc.

Of course, Toa can also convert HTML to Markdown.

Install

pip install toa

Can be used after installation

toa -h
toa-mcp -help

It may only support following content can be converted to HTML.

• such as json, jsonl, txt, md, etc.
• file streams
• external links (such as web crawlers, FTP, and WebSockets)

Extras: [mcp], [csv], [docx], [epub], [excel], [outlook], [pdf], [pptx], [rss], [audio], [video]

uv add 'toa[mcp, csv, excel]'

• [lite] supports capabilities other than [video].
• [all] will install all dependencies. If you don't want to make a choice, this might be the best option, and an additional OCR engine needs to be installed.

with OCR If OCR image recognition is required, you can choose an OCR engine. Such as paddleocr, easyocr, pytesseract, for example:

uv add paddlepaddle paddleocr

Usage

Use in command line

toa md.md

toa md.md -o md.html --wrapper default --mermaid --highlight --formula

Note

If the file does not exist, it may be treated as a string and ultimately returned as an HTML fragment.

• --wrapper: HTML fragment wrapper
• -o: output filename
• --mermaid: mermaid rendering is required.
• --highlight: highlight rendering is required.
• --formula: Requires support for rendering mathematical formulas.

standard input

cat pdf.pdf | toa

cat pdf.pdf | toa --wrapper default -o pdf.html --source-url https://example.com/pdf.pdf --engine default

• --source-url: An accessible file address ensures the page displays correctly.
• --engine: rendering engine name

Important

Some content can only be displayed correctly with server support.

fetch from websocket endpoint

toa ws://examples.com/ws
toa wss://examples.com/wss

fetch from ftp server

toa ftp://ftp.example.com/readme.txt

Use it in Python coding

from toa import Toa

toa = Toa()
result = toa.convert('md.md')
print(result.content)  # <p>md.md</p>

toa-mcp

Toa has built-in MCP capabilities. The corresponding separate command is toa-mcp. toa-mcp is a server based on the MCP (Model Context Protocol) protocol, providing a standardized API interface for the toa command-line tool.

Multi-mode Operation:

• HTTP: Provides a RESTful API interface, suitable for web integration.
• SSE: Server sends events, supports real-time push notifications.
• WebSocket: Bidirectional persistent connection for real-time interaction.
• Stdio: Standard input/output communication, suitable for integration with AI assistants.

endpoint	method	description
/sse	Get/POST	Message data is transmitted in a streaming manner (Server-to-Client).
/messages	GET	Standard message retrieval endpoint (Polling/REST).
/mcp	GET	Streamable HTTP (Model Context Protocol / Long-polling transport).
/ws	GET (Upgrade)	Full-duplex bidirectional streaming (WebSocket connection for real-time interaction).
/ping	GET	Check service availability (Heartbeat).

MCP protocol supports

{
  "jsonrpc": "2.0",
  "method": "tools/list",
  "params": {},
  "id": 1
}

Integrations

There are multiple ways to integrate.

Microservice architecture

prepare docker-compose.yml

# docker-compose.yml
services:
  toa-cli:
    image: sunshmo/toa:latest
    entrypoint: ["toa"]
    command: ["--help"]
    stdin_open: true
    tty: true
  toa-mcp:
    image: sunshmo/toa:latest
    entrypoint: ["toa-mcp", "--transport", "http"]
    ports:
      - "56156:56156"

  # other services

Build and start service

docker-compose up toa-cli
docker-compose up -d toa-mcp

AI assistant integration Such as, Cursor, openclaw, claude, etc.

{
  "mcpServers": {
    "toa": {
      "command": "toa-mcp",
      "args": ["--transport", "stdio"]
    }
  }
}

You can also use the docker command.

{
  "mcpServers": {
    "toa": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-v",
        "./path/data:/data", // resource directory mapping
        "sunshmo/toa:latest",
        "toa-mcp",
        "--transport",
        "stdio"
      ]
    }
  }
}

Integration with coding If toa-mcp is already running, then you can call it in the program. No programming language restrictions. For example, below uses HTTP transport.

async def with_http():
    import httpx

    async with httpx.AsyncClient(follow_redirects=True) as client:
        response = await client.post(
            "http://localhost:56156/mcp",
            headers={"Content-Type": "application/json", "Accept": "application/json"},
            json={"jsonrpc": "2.0", "method": "tools/list", "id": 1},
        )
        print("tools:", response.json())

Below is the capability of Java to call toa-mcp

import com.fasterxml.jackson.databind.JsonNode;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.fasterxml.jackson.databind.node.ObjectNode;

import java.net.URI;
import java.net.http.HttpClient;
import java.net.http.HttpRequest;
import java.net.http.HttpResponse;
import java.time.Duration;

public class McpHttpClient {
    private static final ObjectMapper mapper = new ObjectMapper();
    
    public static void withHttp() throws Exception {
        HttpClient client = HttpClient.newBuilder()
            .followRedirects(HttpClient.Redirect.NORMAL)
            .connectTimeout(Duration.ofSeconds(10))
            .build();
        
        // build JSON-RPC request
        ObjectNode request = mapper.createObjectNode();
        request.put("jsonrpc", "2.0");
        request.put("method", "tools/list");
        request.put("id", 1);
        
        HttpRequest httpRequest = HttpRequest.newBuilder()
            .uri(URI.create("http://localhost:56156/mcp"))
            .header("Content-Type", "application/json")
            .header("Accept", "application/json")
            .timeout(Duration.ofSeconds(30))
            .POST(HttpRequest.BodyPublishers.ofString(mapper.writeValueAsString(request)))
            .build();
        
        HttpResponse<String> response = client.send(httpRequest, HttpResponse.BodyHandlers.ofString());
        
        if (response.statusCode() == 200) {
            JsonNode result = mapper.readTree(response.body());
            System.out.println("tools: " + result.toPrettyString());
        } else {
            System.err.println("HTTP Error: " + response.statusCode());
        }
    }
    
    public static void main(String[] args) throws Exception {
        withHttp();
    }
}

with web MCP Inspector debug

npx @modelcontextprotocol/inspector

The accessible UI address after startup is http://localhost:6274. You can switch the Transport Type to STDIO, SSE, or Streamable HTTP.

STDIO: Enter toa-mcp in the command field and click Connect.

SSE:

• Start service: toa-mcp --transport http, or toa-mcp --transport sse, or toa-mcp --transport ws can all start the web service.
• Enter URL: http://localhost:56156/sse
• click Connect

Streamable HTTP:

• Enter URL: http://localhost:56156/mcp
• click Connect

Docker

Running MCP Web Server Using Docker to build eliminates the hassle of OCR installation. Running toa-mcp service in Docker.

docker run -d \
  -p 56156:56156 \
  --name toa \
  -v /the/host/dir:/data \
  sunshmo/toa:latest \
  toa-mcp --transport http

Verify if the service is usable, visit http://localhost:56156/ping, Seeing "pong" indicates a normal startup.

Tip

-v option maps a host directory to a container directory, allowing you to use a directory path within the container as the target file path when calling the function.

STDIO

docker run --rm -i sunshmo/toa:latest toa < your.eml

Plugin

In reality, there may be files that Toa currently does not support for conversion, or you may want to override Toa's default capabilities. In such cases, you can achieve this by using an extension plugin.

You can do like this:

1. Implement a class that inherits from BaseConverter
2. then register the converter with Toa

Below is one full example scenario.

plugin example

from typing import BinaryIO

from toa import Toa, BaseConverter, InputInfo, OutputInfo, ConversionResult


class MyToaConverter(BaseConverter):
    def __init__(self) -> None:
        super().__init__()

    @property
    def check_dependencies(self) -> bool:
        """
        Returns a boolean value. Returns True if there are no external dependencies;
        otherwise, the conversion will throw an error.
        """
        return True

    @property
    def suffixes(self) -> list[str]:
        """Define extension (Format: ['.toa'])"""
        return [".toa"]

    @property
    def mime_types(self) -> list[str]:
        return []

    def convert(
        self,
        input_stream: BinaryIO,
        input_info: InputInfo,
        output_info: OutputInfo | None = None,
        **kwargs,
    ) -> ConversionResult:
        return ConversionResult(content="Result from plugin conversion")


toa = Toa()

# register
toa.add_converter(MyToaConverter())

result = toa.convert("abc.toa")
print(result.content)

Note

Either suffixes or mime_types must be a value that is not of type NoneType.