JSON Blocks

Fenced JSON

Extract JSON from a fenced code block:

from pydantic import BaseModel
from md2pydantic import MDConverter

class ServerConfig(BaseModel):
    host: str
    port: int
    debug: bool

markdown = '''Here is the config:

```json
{
    "host": "localhost",
    "port": 8080,
    "debug": true
}
```
'''

config = MDConverter(ServerConfig).parse_json(markdown)
# ServerConfig(host='localhost', port=8080, debug=True)

Note

parse_json returns a single model instance, not a list. It tries each JSON block in document order and returns the first one that validates.

Inline JSON

md2pydantic also detects inline (unfenced) JSON objects:

markdown = """
The server config is {"host": "localhost", "port": 8080,
"debug": true} as specified.
"""

config = MDConverter(ServerConfig).parse_json(markdown)

Recovery Features

LLM output often contains malformed JSON. md2pydantic attempts to fix these issues automatically:

Issue	Example	Recovery
Trailing commas	{"a": 1,}	Comma removed
Single quotes	{'a': 1}	Replaced with "
Unquoted keys	{host: "localhost"}	Keys quoted
Truncated JSON	{"a": 1, "b": 2	Brackets closed

Tip

Recovery is best-effort. If the JSON is too malformed to salvage, an ExtractionError is raised.