litserve Implementation of Microsoft markitdown

This exposes markitdown via API on port 8000 at /convert. The docker image is published at mathematicalmichael/md-api:latest docker-hub

Usage

Make sure you have installed uv: pip install uv first, or modify the makefile appropriately for your environment. uv is used to resolve and install dependencies. It creates the requirements.txt file required by docker.

make run will build and run the docker image, but the command is

docker build -t doc-api .

docker run --rm -p 8000:8000 doc-api

Alternatively, you can start the server locally (without docker):

uv run server.py

Example client.py file:

Note: once the server is up, you can run make docker-test to run the following script against a local endpoint using the same image. Alternatively, make test will use uv to run a local test.

import argparse
import json
from datetime import datetime
from pathlib import Path

import requests

API_URL = "http://127.0.0.1:8000/convert"


def send_request(path):
    try:
        with open(path, "rb") as inputFile:
            # Use a tuple (filename, file object) for 'files'
            response = requests.post(
                API_URL, files={"content": (Path(path).name, inputFile)}
            )
    except FileNotFoundError:
        # passing a URL for markdownit to scrape
        response = requests.post(API_URL, json={"content": path})
    if response.status_code == 200:
        output = json.loads(response.content)["output"]
        save_output(output)
    else:
        print(
            f"Error: Response with status code {response.status_code} - {response.text}"
        )


def save_output(output):
    timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S").lower()
    Path("output").mkdir(exist_ok=True)
    filename = f"output/{timestamp}.md"

    with open(filename, "wb") as output_file:
        output = output.encode("utf-8")
        output_file.write(output)

    print(f"File saved to {filename}")


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description="Send file to markdownit server and receive generated markdown content."
    )
    parser.add_argument("--path", required=True, help="Path for the document file.")
    args = parser.parse_args()

    send_request(args.path)

For local data:

curl -X POST -F "content=@tests/test_rss.xml;filename=test_rss.xml" http://127.0.0.1:8000/convert

For remote data:

curl --request POST \
    --url http://0.0.0.0:8000/convert \
    --header 'Content-Type: application/json' \
    --data '{
        "content": "https://wikipedia.org"
    }'

Contributing

Contributions are welcome!

pip install uv and the rest should be straightforward - consult the makefile for more details.